c F&A

 

The questions answered here are divided into several categories:

	 1. Declarations and Initializations
	 2. Structures, Unions, and Enumerations
	 3. Expressions
	 4. Pointers
	 5. Null Pointers
	 6. Arrays and Pointers
	 7. Memory Allocation
	 8. Characters and Strings
	 9. Boolean Expressions and Variables
	10. C Preprocessor
	11. ANSI/ISO Standard C
	12. Stdio
	13. Library Functions
	14. Floating Point
	15. Variable-Length Argument Lists
	16. Strange Problems
	17. Style
	18. Tools and Resources
	19. System Dependencies
	20. Miscellaneous
	    Bibliography
	    Acknowledgements

(The question numbers within each section are not always continuous,
because they are aligned with the aforementioned book-length version,
which contains even more questions.)

Herewith, some frequently-asked questions and their answers:


Section 1. Declarations and Initializations

1.1:	How should I decide which integer type to use?

A:	If you might need large values (above 32,767 or below -32,767),
	use long.  Otherwise, if space is very important (i.e. if there
	are large arrays or many structures), use short.  Otherwise, use
	int.  If well-defined overflow characteristics are important and
	negative values are not, or if you want to steer clear of sign-
	extension problems when manipulating bits or bytes, use one of
	the corresponding unsigned types.  (Beware when mixing signed
	and unsigned values in expressions, though.)

	Although character types (especially unsigned char) can be used
	as "tiny" integers, doing so is sometimes more trouble than it's
	worth, due to unpredictable sign extension and increased code
	size.  (Using unsigned char can help; see question 12.1 for a
	related problem.)

	A similar space/time tradeoff applies when deciding between
	float and double.  None of the above rules apply if pointers to
	the variable must have a particular type.

	If for some reason you need to declare something with an *exact*
	size (usually the only good reason for doing so is when
	attempting to conform to some externally-imposed storage layout,
	but see question 20.5), be sure to encapsulate the choice behind
	an appropriate typedef, such as those in C99's .

	If you need to manipulate huge values, larger than the
	guaranteed range of C's built-in types, see question 18.15d.

	References: K&R1 Sec. 2.2 p. 34; K&R2 Sec. 2.2 p. 36, Sec. A4.2
	pp. 195-6, Sec. B11 p. 257; ISO Sec. 5.2.4.2.1, Sec. 6.1.2.5;
	H&S Secs. 5.1,5.2 pp. 110-114.

1.4:	What should the 64-bit type be on a machine that can support it?

A:	The new C99 Standard specifies type long long as effectively
	being at least 64 bits, and this type has been implemented by a
	number of compilers for some time.  (Others have implemented
	extensions such as __longlong.)  On the other hand, it's also
	appropriate to implement type short int as 16, int as 32, and
	long int as 64 bits, and some compilers do.

	See also question 18.15d.

	References: C9X Sec. 5.2.4.2.1, Sec. 6.1.2.5.

1.7:	What's the best way to declare and define global variables
	and functions?

A:	First, though there can be many "declarations" (and in many
	translation units) of a single global variable or function,
	there must be exactly one "definition", where the definition is
	the declaration that actually allocates space, and provides an
	initialization value, if any.  The best arrangement is to place
	each definition in some relevant .c file, with an external
	declaration in a header (".h") file, which is included wherever
	the declaration is needed.  The .c file containing the
	definition should also #include the same header file, so the
	compiler can check that the definition matches the declarations.

	This rule promotes a high degree of portability: it is
	consistent with the requirements of the ANSI C Standard, and is
	also consistent with most pre-ANSI compilers and linkers.  (Unix
	compilers and linkers typically use a "common model" which
	allows multiple definitions, as long as at most one is
	initialized; this behavior is mentioned as a "common extension"
	by the ANSI Standard, no pun intended.)

	It is possible to use preprocessor tricks to arrange that a line
	like

		DEFINE(int, i);

	need only be entered once in one header file, and turned into a
	definition or a declaration depending on the setting of some
	macro, but it's not clear if this is worth the trouble.

	It's especially important to put global declarations in header
	files if you want the compiler to catch inconsistent
	declarations for you.  In particular, never place a prototype
	for an external function in a .c file: it wouldn't generally be
	checked for consistency with the definition, and an incompatible
	prototype is worse than useless.

	See also questions 10.6 and 18.8.

	References: K&R1 Sec. 4.5 pp. 76-7; K&R2 Sec. 4.4 pp. 80-1; ISO
	Sec. 6.1.2.2, Sec. 6.7, Sec. 6.7.2, Sec. G.5.11; Rationale
	Sec. 3.1.2.2; H&S Sec. 4.8 pp. 101-104, Sec. 9.2.3 p. 267; CT&P
	Sec. 4.2 pp. 54-56.

1.11:	What does extern mean in a function declaration?

A:	It can be used as a stylistic hint to indicate that the
	function's definition is probably in another source file, but
	there is no formal difference between

		extern int f();

	and

		int f();

	References: ISO Sec. 6.1.2.2, Sec. 6.5.1; Rationale
	Sec. 3.1.2.2; H&S Secs. 4.3,4.3.1 pp. 75-6.

1.12:	What's the auto keyword good for?

A:	Nothing; it's archaic.  See also question 20.37.

	References: K&R1 Sec. A8.1 p. 193; ISO Sec. 6.1.2.4, Sec. 6.5.1;
	H&S Sec. 4.3 p. 75, Sec. 4.3.1 p. 76.

1.14:	I can't seem to define a linked list successfully.  I tried

		typedef struct {
			char *item;
			NODEPTR next;
		} *NODEPTR;

	but the compiler gave me error messages.  Can't a structure in C
	contain a pointer to itself?

A:	Structures in C can certainly contain pointers to themselves;
	the discussion and example in section 6.5 of K&R make this
	clear.  The problem with the NODEPTR example is that the typedef
	has not yet been defined at the point where the "next" field is
	declared.  To fix this code, first give the structure a tag
	(e.g. "struct node").  Then, declare the "next" field as a
	simple "struct node *", or disentangle the typedef declaration
	from the structure definition, or both.  One corrected version
	would be

		struct node {
			char *item;
			struct node *next;
		};

		typedef struct node *NODEPTR;

	and there are at least three other equivalently correct ways of
	arranging it.

	A similar problem, with a similar solution, can arise when
	attempting to declare a pair of typedef'ed mutually referential
	structures.

	See also question 2.1.

	References: K&R1 Sec. 6.5 p. 101; K&R2 Sec. 6.5 p. 139; ISO
	Sec. 6.5.2, Sec. 6.5.2.3; H&S Sec. 5.6.1 pp. 132-3.

1.21:	How do I construct and understand declarations of complicated
	types such as "array of N pointers to functions returning
	pointers to functions returning pointers to char"?

A:	There are at least three ways of answering this question:

	1.  char *(*(*a[N])())();

	2.  Build the declaration up incrementally, using typedefs:

		typedef char *pc;	/* pointer to char */
		typedef pc fpc();	/* function returning pointer to char */
		typedef fpc *pfpc;	/* pointer to above */
		typedef pfpc fpfpc();	/* function returning... */
		typedef fpfpc *pfpfpc;	/* pointer to... */
		pfpfpc a[N];		/* array of... */

	3.  Use the cdecl program, which turns English into C and vice
	    versa:

		cdecl> declare a as array of pointer to function returning
			pointer to function returning pointer to char
		char *(*(*a[])())()

	    cdecl can also explain complicated declarations, help with
	    casts, and indicate which set of parentheses the parameters
	    go in (for complicated function definitions, like the one
	    above).  See question 18.1.

	A good book on C should explain how to read these complicated
	declarations "inside out" to understand them ("declaration
	mimics use").

	The pointer-to-function declarations in the examples above have
	not included parameter type information.  When the parameters
	have complicated types, declarations can *really* get messy.
	(Modern versions of cdecl can help here, too.)

	References: K&R2 Sec. 5.12 p. 122; ISO Sec. 6.5ff (esp.
	Sec. 6.5.4); H&S Sec. 4.5 pp. 85-92, Sec. 5.10.1 pp. 149-50.

1.25:	My compiler is complaining about an invalid redeclaration of a
	function, but I only define it once and call it once.

A:	Functions which are called without a declaration in scope,
	perhaps because the first call precedes the function's
	definition, are assumed to be declared as returning int (and
	without any argument type information), leading to discrepancies
	if the function is later declared or defined otherwise.  All
	functions should be (and non-int functions must be) declared
	before they are called.

	Another possible source of this problem is that the function has
	the same name as another one declared in some header file.

	See also questions 11.3 and 15.1.

	References: K&R1 Sec. 4.2 p. 70; K&R2 Sec. 4.2 p. 72; ISO
	Sec. 6.3.2.2; H&S Sec. 4.7 p. 101.

1.25b:	What's the right declaration for main()?
	Is void main() correct?

A:	See questions 11.12a through 11.15.  (But no, it's not correct.)

1.30:	What am I allowed to assume about the initial values of
	variables and arrays which are not explicitly initialized?
	If global variables start out as "zero", is that good enough
	for null pointers and floating-point zeroes?

A:	Uninitialized variables with "static" duration (that is, those
	declared outside of functions, and those declared with the
	storage class static), are guaranteed to start out as zero, just
	as if the programmer had typed "= 0".  Therefore, such variables
	are implicitly initialized to the null pointer (of the correct
	type; see also section 5) if they are pointers, and to 0.0 if
	they are floating-point.

	Variables with "automatic" duration (i.e. local variables
	without the static storage class) start out containing garbage,
	unless they are explicitly initialized.  (Nothing useful can be
	predicted about the garbage.)

	These rules do apply to arrays and structures (termed
	"aggregates"); arrays and structures are considered "variables"
	as far as initialization is concerned.

	Dynamically-allocated memory obtained with malloc() and
	realloc() is likely to contain garbage, and must be initialized
	by the calling program, as appropriate.  Memory obtained with
	calloc() is all-bits-0, but this is not necessarily useful for
	pointer or floating-point values (see question 7.31, and section
	5).

	References: K&R1 Sec. 4.9 pp. 82-4; K&R2 Sec. 4.9 pp. 85-86; ISO
	Sec. 6.5.7, Sec. 7.10.3.1, Sec. 7.10.5.3; H&S Sec. 4.2.8 pp.
	72-3, Sec. 4.6 pp. 92-3, Sec. 4.6.2 pp. 94-5, Sec. 4.6.3 p. 96,
	Sec. 16.1 p. 386.

1.31:	This code, straight out of a book, isn't compiling:

		int f()
		{
			char a[] = "Hello, world!";
		}

A:	Perhaps you have an old, pre-ANSI compiler, which doesn't allow
	initialization of "automatic aggregates" (i.e. non-static local
	arrays, structures, or unions).  See also question 11.29.

1.31b:	What's wrong with this initialization?

		char *p = malloc(10);

	My compiler is complaining about an "invalid initializer",
	or something.

A:	Is the declaration of a static or non-local variable?  Function
	calls are allowed in initializers only for automatic variables
	(that is, for local, non-static variables).

1.32:	What is the difference between these initializations?

		char a[] = "string literal";
		char *p  = "string literal";

	My program crashes if I try to assign a new value to p[i].

A:	A string literal can be used in two slightly different ways.  As
	an array initializer (as in the declaration of char a[] in the
	question), it specifies the initial values of the characters in
	that array.  Anywhere else, it turns into an unnamed, static
	array of characters, which may be stored in read-only memory,
	and which therefore cannot necessarily be modified.  In an
	expression context, the array is converted at once to a pointer,
	as usual (see section 6), so the second declaration initializes
	p to point to the unnamed array's first element.

	(For compiling old code, some compilers have a switch
	controlling whether string literals are writable or not.)

	See also questions 1.31, 6.1, 6.2, 6.8, and 11.8b.

	References: K&R2 Sec. 5.5 p. 104; ISO Sec. 6.1.4, Sec. 6.5.7;
	Rationale Sec. 3.1.4; H&S Sec. 2.7.4 pp. 31-2.

1.34:	I finally figured out the syntax for declaring pointers to
	functions, but now how do I initialize one?

A:	Use something like

		extern int func();
		int (*fp)() = func;

	When the name of a function appears in an expression, it
	"decays" into a pointer (that is, it has its address implicitly
	taken), much as an array name does.

	A prior, explicit declaration for the function (perhaps in a
	header file) is normally needed.  The implicit external function
	declaration that can occur when a function is called does not
	help when a function name's only use is for its value.

	See also questions 1.25 and 4.12.


Section 2. Structures, Unions, and Enumerations

2.1:	What's the difference between these two declarations?

		struct x1 { ... };
		typedef struct { ... } x2;

A:	The first form declares a "structure tag"; the second declares a
	"typedef".  The main difference is that you subsequently refer
	to the first type as "struct x1" and the second simply as "x2".
	That is, the second declaration is of a slightly more abstract
	type -- its users don't necessarily know that it is a structure,
	and the keyword struct is not used when declaring instances of it.

2.2:	Why doesn't

		struct x { ... };
		x thestruct;

	work?

A:	C is not C++.  Typedef names are not automatically generated for
	structure tags.  See also questions 1.14 and 2.1.

2.3:	Can a structure contain a pointer to itself?

A:	Most certainly.  See also question 1.14.

2.4:	How can I implement opaque (abstract) data types in C?

A:	One good way is for clients to use structure pointers (perhaps
	additionally hidden behind typedefs) which point to structure
	types which are not publicly defined.  It's legal to declare
	and use "anonymous" structure pointers (that is, pointers to
	structures of incomplete type), as long as no attempt is made to
	access the members -- which of course is exactly the point of an
	opaque type.

2.4b:	Is there a good way of simulating OOP-style inheritance, or
	other OOP features, in C?

A:	It's straightforward to implement simple "methods" by placing
	function pointers in structures.  You can make various clumsy,
	brute-force attempts at inheritance using the preprocessor or by
	having structures contain "base types" as initial subsets, but
	it won't be perfect.  There's obviously no operator overloading,
	and overriding (i.e. of "methods" in "derived classes") would
	have to be done by hand.

	Obviously, if you need "real" OOP, you'll want to use a language
	that supports it, such as C++.

2.6:	I came across some code that declared a structure like this:

		struct name {
			int namelen;
			char namestr[1];
		};

	and then did some tricky allocation to make the namestr array
	act like it had several elements.  Is this legal or portable?

A:	This technique is popular, although Dennis Ritchie has called it
	"unwarranted chumminess with the C implementation."  An official
	interpretation has deemed that it is not strictly conforming
	with the C Standard, although it does seem to work under all
	known implementations.  (Compilers which check array bounds
	carefully might issue warnings.)

	Another possibility is to declare the variable-size element very
	large, rather than very small; in the case of the above example:

		...
		char namestr[MAXSIZE];

	where MAXSIZE is larger than any name which will be stored.
	However, it looks like this technique is disallowed by a strict
	interpretation of the Standard as well.  Furthermore, either of
	these "chummy" structures must be used with care, since the
	programmer knows more about their size than the compiler does.

	C99 introduces the concept of a "flexible array member", which
	allows the size of an array to be omitted if it is the last
	member in a structure, thus providing a well-defined solution.

	References: Rationale Sec. 3.5.4.2; C9X Sec. 6.5.2.1.

2.8:	Is there a way to compare structures automatically?

A:	No.  There is not a good way for a compiler to implement
	structure comparison (i.e. to support the == operator for
	structures) which is consistent with C's low-level flavor.
	A simple byte-by-byte comparison could founder on random bits
	present in unused "holes" in the structure (see question 2.12).
	A field-by-field comparison might require unacceptable amounts
	of repetitive code for large structures.

	If you need to compare two structures, you'll have to write your
	own function to do so, field by field.

	References: K&R2 Sec. 6.2 p. 129; Rationale Sec. 3.3.9; H&S
	Sec. 5.6.2 p. 133.

2.10:	How can I pass constant values to functions which accept
	structure arguments?

A:	Traditional C had no way of generating anonymous structure
	values; you had to use a temporary structure variable or a
	little structure-building function.

	C99 introduces "compound literals", one form of which provides
	for structure constants.  For example, to pass a constant
	coordinate pair to a hypothetical plotpoint() function which
	expects a struct point, you can call

		plotpoint((struct point){1, 2});

	Combined with "designated initializers" (another C99 feature),
	it is also possible to specify member values by name:

		plotpoint((struct point){.x=1, .y=2});

	See also question 4.10.

	References: C9X Sec. 6.3.2.5, Sec. 6.5.8.

2.11:	How can I read/write structures from/to data files?

A:	It is relatively straightforward to write a structure out using
	fwrite():

		fwrite(&somestruct, sizeof somestruct, 1, fp);

	and a corresponding fread invocation can read it back in.
	However, data files so written will *not* be portable (see
	questions 2.12 and 20.5).  Also, if the structure contains any
	pointers, only the pointer values will be written, and they are
	most unlikely to be valid when read back in.  Finally, note that
	for widespread portability you must use the "b" flag when
	opening the files; see question 12.38.

	A more portable solution, though it's a bit more work initially,
	is to write a pair of functions for writing and reading a
	structure, field-by-field, in a portable (perhaps even human-
	readable) way.

	References: H&S Sec. 15.13 p. 381.

2.12:	My compiler is leaving holes in structures, which is wasting
	space and preventing "binary" I/O to external data files.  Why?
	Can I turn this off, or otherwise control the alignment of
	structure fields?

A:	Those "holes" provide "padding", which may be needed in order to
	preserve the "alignment" of later fields of the structure.  For
	efficient access, most processors prefer (or require) that
	multibyte objects (e.g. structure members of any type larger
	than char) not sit at arbitrary memory addresses, but rather at
	addresses which are multiples of 2 or 4 or the object size.

	Your compiler may provide an extension to give you explicit
	control over struct alignment (perhaps involving a #pragma; see
	question 11.20), but there is no standard method.

	See also question 20.5.

	References: K&R2 Sec. 6.4 p. 138; H&S Sec. 5.6.4 p. 135.

2.13:	Why does sizeof report a larger size than I expect for a
	structure type, as if there were padding at the end?

A:	Padding at the end of a structure may be necessary to preserve
	alignment when an array of contiguous structures is allocated.
	Even when the structure is not part of an array, the padding
	remains, so that sizeof can always return a consistent size.
	See also question 2.12 above.

	References: H&S Sec. 5.6.7 pp. 139-40.

2.14:	How can I determine the byte offset of a field within a
	structure?

A:	ANSI C defines the offsetof() macro in , which lets
	you compute the offset of field f in struct s as
	offsetof(struct s, f).  If for some reason you have to code this
	sort of thing yourself, one possibility is

		#define offsetof(type, f) ((size_t) /
			((char *)&((type *)0)->f - (char *)(type *)0))

	This implementation is not 100% portable; some compilers may
	legitimately refuse to accept it.

	References: ISO Sec. 7.1.6; Rationale Sec. 3.5.4.2; H&S
	Sec. 11.1 pp. 292-3.

2.15:	How can I access structure fields by name at run time?

A:	Keep track of the field offsets as computed using the offsetof()
	macro (see question 2.14).  If structp is a pointer to an
	instance of the structure, and field f is an int having offset
	offsetf, f's value can be set indirectly with

		*(int *)((char *)structp + offsetf) = value;

2.18:	This program works correctly, but it dumps core after it
	finishes.  Why?

		struct list {
			char *item;
			struct list *next;
		}

		/* Here is the main program. */

		main(argc, argv)
		{ ... }

A:	A missing semicolon causes main() to be declared as returning a
	structure.  (The connection is hard to see because of the
	intervening comment.)  Since structure-valued functions are
	usually implemented by adding a hidden return pointer, the
	generated code for main() tries to accept three arguments,
	although only two are passed (in this case, by the C start-up
	code).  See also questions 10.9 and 16.4.

	References: CT&P Sec. 2.3 pp. 21-2.

2.20:	Can I initialize unions?

A:	In the original ANSI C, an initializer was allowed only for the
	first-named member of a union.  C99 introduces "designated
	initializers" which can be used to initialize any member.

	References: K&R2 Sec. 6.8 pp. 148-9; ISO Sec. 6.5.7; C9X
	Sec. 6.5.8; H&S Sec. 4.6.7 p. 100.

2.22:	What's the difference between an enumeration and a set of
	preprocessor #defines?

A:	There is little difference.  The C Standard says that
	enumerations may be freely intermixed with other integral types,
	without errors.  (If, on the other hand, such intermixing were
	disallowed without explicit casts, judicious use of enumerations
	could catch certain programming errors.)

	Some advantages of enumerations are that the numeric values are
	automatically assigned, that a debugger may be able to display
	the symbolic values when enumeration variables are examined, and
	that they obey block scope.  (A compiler may also generate
	nonfatal warnings when enumerations are indiscriminately mixed,
	since doing so can still be considered bad style.)  A
	disadvantage is that the programmer has little control over
	those nonfatal warnings; some programmers also resent not having
	control over the sizes of enumeration variables.

	References: K&R2 Sec. 2.3 p. 39, Sec. A4.2 p. 196; ISO
	Sec. 6.1.2.5, Sec. 6.5.2, Sec. 6.5.2.2, Annex F; H&S Sec. 5.5
	pp. 127-9, Sec. 5.11.2 p. 153.

2.24:	Is there an easy way to print enumeration values symbolically?

A:	No.  You can write a little function to map an enumeration
	constant to a string.  (For debugging purposes, a good debugger
	should automatically print enumeration constants symbolically.)


Section 3. Expressions

3.1:	Why doesn't this code:

		a[i] = i++;

	work?

A:	The subexpression i++ causes a side effect -- it modifies i's
	value -- which leads to undefined behavior since i is also
	referenced elsewhere in the same expression, and there's no way
	to determine whether the reference (in a[i] on the left-hand
	side) should be to the old or the new value.  (Note that
	although the language in K&R suggests that the behavior of this
	expression is unspecified, the C Standard makes the stronger
	statement that it is undefined -- see question 11.33.)

	References: K&R1 Sec. 2.12; K&R2 Sec. 2.12; ISO Sec. 6.3; H&S
	Sec. 7.12 pp. 227-9.

3.2:	Under my compiler, the code

		int i = 7;
		printf("%d/n", i++ * i++);

	prints 49.  Regardless of the order of evaluation, shouldn't it
	print 56?

A:	Although the postincrement and postdecrement operators ++ and --
	perform their operations after yielding the former value, the
	implication of "after" is often misunderstood.  It is *not*
	guaranteed that an increment or decrement is performed
	immediately after giving up the previous value and before any
	other part of the expression is evaluated.  It is merely
	guaranteed that the update will be performed sometime before the
	expression is considered "finished" (before the next "sequence
	point," in ANSI C's terminology; see question 3.8).  In the
	example, the compiler chose to multiply the previous value by
	itself and to perform both increments later.

	The behavior of code which contains multiple, ambiguous side
	effects has always been undefined.  (Loosely speaking, by
	"multiple, ambiguous side effects" we mean any combination of
	increment, decrement, and assignment operators in a single
	expression which causes the same object either to be modified
	twice or modified and then inspected.  This is a rough
	definition; see question 3.8 for a precise one, and question
	11.33 for the meaning of "undefined.")  Don't even try to find
	out how your compiler implements such things (contrary to the
	ill-advised exercises in many C textbooks); as K&R wisely point
	out, "if you don't know *how* they are done on various machines,
	that innocence may help to protect you."

	References: K&R1 Sec. 2.12 p. 50; K&R2 Sec. 2.12 p. 54; ISO
	Sec. 6.3; H&S Sec. 7.12 pp. 227-9; CT&P Sec. 3.7 p. 47; PCS
	Sec. 9.5 pp. 120-1.

3.3:	I've experimented with the code

		int i = 3;
		i = i++;

	on several compilers.  Some gave i the value 3, and some gave 4.
	Which compiler is correct?

A:	There is no correct answer; the expression is undefined.  See
	questions 3.1, 3.8, 3.9, and 11.33.  (Also, note that neither
	i++ nor ++i is the same as i+1.  If you want to increment i,
	use i=i+1, i+=1, i++, or ++i, not some combination.  See also
	question 3.12b.)

3.3b:	Here's a slick expression:

		a ^= b ^= a ^= b

	It swaps a and b without using a temporary.

A:	Not portably, it doesn't.  It attempts to modify the variable a
	twice between sequence points, so its behavior is undefined.

	For example, it has been reported that when given the code

		int a = 123, b = 7654;
		a ^= b ^= a ^= b;

	the SCO Optimizing C compiler (icc) sets b to 123 and a to 0.

	See also questions 3.1, 3.8, 10.3, and 20.15c.

3.4:	Can I use explicit parentheses to force the order of evaluation
	I want?  Even if I don't, doesn't precedence dictate it?

A:	Not in general.

	Operator precedence and explicit parentheses impose only a
	partial ordering on the evaluation of an expression.  In the
	expression

		f() + g() * h()

	although we know that the multiplication will happen before the
	addition, there is no telling which of the three functions will
	be called first.

	When you need to ensure the order of subexpression evaluation,
	you may need to use explicit temporary variables and separate
	statements.

	References: K&R1 Sec. 2.12 p. 49, Sec. A.7 p. 185; K&R2
	Sec. 2.12 pp. 52-3, Sec. A.7 p. 200.

3.5:	But what about the && and || operators?
	I see code like "while((c = getchar()) != EOF && c != '/n')" ...

A:	There is a special "short-circuiting" exception for these
	operators: the right-hand side is not evaluated if the left-hand
	side determines the outcome (i.e. is true for || or false for
	&&).  Therefore, left-to-right evaluation is guaranteed, as it
	also is for the comma operator.  Furthermore, all of these
	operators (along with ?:) introduce an extra internal sequence
	point (see question 3.8).

	References: K&R1 Sec. 2.6 p. 38, Secs. A7.11-12 pp. 190-1; K&R2
	Sec. 2.6 p. 41, Secs. A7.14-15 pp. 207-8; ISO Sec. 6.3.13,
	Sec. 6.3.14, Sec. 6.3.15; H&S Sec. 7.7 pp. 217-8, Sec. 7.8 pp.
	218-20, Sec. 7.12.1 p. 229; CT&P Sec. 3.7 pp. 46-7.

3.8:	How can I understand these complex expressions?  What's a
	"sequence point"?

A:	A sequence point is a point in time (at the end of the
	evaluation of a full expression, or at the ||, &&, ?:, or comma
	operators, or just before a function call) at which the dust
	has settled and all side effects are guaranteed to be complete.
	The ANSI/ISO C Standard states that

		Between the previous and next sequence point an
		object shall have its stored value modified at
		most once by the evaluation of an expression.
		Furthermore, the prior value shall be accessed
		only to determine the value to be stored.

	The second sentence can be difficult to understand.  It says
	that if an object is written to within a full expression, any
	and all accesses to it within the same expression must be
	directly involved in the computation of the value to be written.
	This rule effectively constrains legal expressions to those in
	which the accesses demonstrably precede the modification.  For
	example, i = i + 1 is legal, but not a[i] = i++ (see question
	3.1).

	See also question 3.9 below.

	References: ISO Sec. 5.1.2.3, Sec. 6.3, Sec. 6.6, Annex C;
	Rationale Sec. 2.1.2.3; H&S Sec. 7.12.1 pp. 228-9.

3.9:	So given

		a[i] = i++;

	we don't know which cell of a[] gets written to, but i does get
	incremented by one, right?

A:	Not necessarily!  Once an expression or program becomes
	undefined, *all* aspects of it become undefined.  See questions
	3.2, 3.3, 11.33, and 11.35.

3.12a:	What's the difference between ++i and i++?

A:	If your C book doesn't explain, get a better one.  Briefly:
	++i adds one to the stored value of i and "returns" the new,
	incremented value to the surrounding expression; i++ adds one
	to i but returns the prior, unincremented value.

3.12b:	If I'm not using the value of the expression, should I use ++i
	or i++ to increment a variable?

A:	Since the two forms differ only in the value yielded, they are
	entirely equivalent when only their side effect is needed.
	(However, the prefix form is preferred in C++.)  See also
	question 3.3.

	References: K&R1 Sec. 2.8 p. 43; K&R2 Sec. 2.8 p. 47; ISO
	Sec. 6.3.2.4, Sec. 6.3.3.1; H&S Sec. 7.4.4 pp. 192-3, Sec. 7.5.8
	pp. 199-200.

3.14:	Why doesn't the code

		int a = 1000, b = 1000;
		long int c = a * b;

	work?

A:	Under C's integral promotion rules, the multiplication is
	carried out using int arithmetic, and the result may overflow or
	be truncated before being promoted and assigned to the long int
	left-hand side.  Use an explicit cast to force long arithmetic:

		long int c = (long int)a * b;

	Notice that (long int)(a * b) would *not* have the desired
	effect.

	A similar problem can arise when two integers are divided, with
	the result assigned to a floating-point variable; the solution
	is similar, too.

	References: K&R1 Sec. 2.7 p. 41; K&R2 Sec. 2.7 p. 44; ISO
	Sec. 6.2.1.5; H&S Sec. 6.3.4 p. 176; CT&P Sec. 3.9 pp. 49-50.

3.16:	I have a complicated expression which I have to assign to one of
	two variables, depending on a condition.  Can I use code like
	this?

		((condition) ? a : b) = complicated_expression;

A:	No.  The ?: operator, like most operators, yields a value, and
	you can't assign to a value.  (In other words, ?: does not yield
	an "lvalue".)  If you really want to, you can try something like

		*((condition) ? &a : &b) = complicated_expression;

	although this is admittedly not as pretty.

	References: ISO Sec. 6.3.15; H&S Sec. 7.1 pp. 179-180.


Section 4. Pointers

4.2:	I'm trying to declare a pointer and allocate some space for it,
	but it's not working.  What's wrong with this code?

		char *p;
		*p = malloc(10);

A:	The pointer you declared is p, not *p.  When you're manipulating
	the pointer itself (for example when you're setting it to make
	it point somewhere), you just use the name of the pointer:

		p = malloc(10);

	It's when you're manipulating the pointed-to memory that you use
	* as an indirection operator:

		*p = 'H';

	See also questions 1.21, 7.1, 7.3c, and 8.3.

	References: CT&P Sec. 3.1 p. 28.

4.3:	Does *p++ increment p, or what it points to?

A:	The postfix ++ and -- operators essentially have higher
	precedence than the prefix unary operators.  Therefore, *p++ is
	equivalent to *(p++); it increments p, and returns the value
	which p pointed to before p was incremented.  To increment the
	value pointed to by p, use (*p)++ (or perhaps ++*p, if the order
	of the side effect doesn't matter).

	References: K&R1 Sec. 5.1 p. 91; K&R2 Sec. 5.1 p. 95; ISO
	Sec. 6.3.2, Sec. 6.3.3; H&S Sec. 7.4.4 pp. 192-3, Sec. 7.5 p.
	193, Secs. 7.5.7,7.5.8 pp. 199-200.

4.5:	I have a char * pointer that happens to point to some ints, and
	I want to step it over them.  Why doesn't

		((int *)p)++;

	work?

A:	In C, a cast operator does not mean "pretend these bits have a
	different type, and treat them accordingly"; it is a conversion
	operator, and by definition it yields an rvalue, which cannot be
	assigned to, or incremented with ++.  (It is either an accident
	or a deliberate but nonstandard extension if a particular
	compiler accepts expressions such as the above.)  Say what you
	mean: use

		p = (char *)((int *)p + 1);

	or (since p is a char *) simply

		p += sizeof(int);

	When possible, however, you should choose appropriate pointer
	types in the first place, rather than trying to treat one type
	as another.

	References: K&R2 Sec. A7.5 p. 205; ISO Sec. 6.3.4; Rationale
	Sec. 3.3.2.4; H&S Sec. 7.1 pp. 179-80.

4.8:	I have a function which accepts, and is supposed to initialize,
	a pointer:

		void f(int *ip)
		{
			static int dummy = 5;
			ip = &dummy;
		}

	But when I call it like this:

		int *ip;
		f(ip);

	the pointer in the caller remains unchanged.

A:	Are you sure the function initialized what you thought it did?
	Remember that arguments in C are passed by value.  The called
	function altered only the passed copy of the pointer.  You'll
	either want to pass the address of the pointer (the function
	will end up accepting a pointer-to-a-pointer), or have the
	function return the pointer.

	See also questions 4.9 and 4.11.

4.9:	Can I use a void ** pointer as a parameter so that a function
	can accept a generic pointer by reference?

A:	Not portably.  There is no generic pointer-to-pointer type in C.
	void * acts as a generic pointer only because conversions (if
	necessary) are applied automatically when other pointer types
	are assigned to and from void *'s; these conversions cannot be
	performed (the correct underlying pointer type is not known) if
	an attempt is made to indirect upon a void ** value which points
	at a pointer type other than void *.

4.10:	I have a function

		extern int f(int *);

	which accepts a pointer to an int.  How can I pass a constant by
	reference?  A call like

		f(&5);

	doesn't seem to work.

A:	In C99, you can use a "compound literal":

		f((int[]){5});

	Prior to C99, you couldn't do this directly; you had to declare
	a temporary variable, and then pass its address to the function:

		int five = 5;
		f(&five);

	See also questions 2.10, 4.8, and 20.1.

4.11:	Does C even have "pass by reference"?

A:	Not really.

	Strictly speaking, C always uses pass by value.  You can
	simulate pass by reference yourself, by defining functions which
	accept pointers and then using the & operator when calling, and
	the compiler will essentially simulate it for you when you pass
	an array to a function (by passing a pointer instead, see
	question 6.4 et al.).  However, C has nothing truly equivalent
	to formal pass by reference or C++ reference parameters.  (On
	the other hand, function-like preprocessor macros can provide a
	form of "pass by name".)

	See also questions 4.8 and 20.1.

	References: K&R1 Sec. 1.8 pp. 24-5, Sec. 5.2 pp. 91-3; K&R2
	Sec. 1.8 pp. 27-8, Sec. 5.2 pp. 95-7; ISO Sec. 6.3.2.2; H&S
	Sec. 9.5 pp. 273-4.

4.12:	I've seen different syntax used for calling functions via
	pointers.  What's the story?

A:	Originally, a pointer to a function had to be "turned into" a
	"real" function, with the * operator (and an extra pair of
	parentheses, to keep the precedence straight), before calling:

		int r, func(), (*fp)() = func;
		r = (*fp)();

	It can also be argued that functions are always called via
	pointers, and that "real" function names always decay implicitly
	into pointers (in expressions, as they do in initializations;
	see question 1.34).  This reasoning means that

		r = fp();

	is legal and works correctly, whether fp is the name of a
	function or a pointer to one.  (The usage has always been
	unambiguous; there is nothing you ever could have done with a
	function pointer followed by an argument list except call the
	function pointed to.)

	The ANSI C Standard essentially adopts the latter
	interpretation, meaning that the explicit * is not required,
	though it is still allowed.

	See also question 1.34.

	References: K&R1 Sec. 5.12 p. 116; K&R2 Sec. 5.11 p. 120; ISO
	Sec. 6.3.2.2; Rationale Sec. 3.3.2.2; H&S Sec. 5.8 p. 147,
	Sec. 7.4.3 p. 190.

4.15:	How do I convert an int to a char *?  I tried a cast, but it's
	not working.

A:	It depends on what you're trying to do.  If you tried a cast
	but it's not working, you're probably trying to convert an
	integer to a string, in which case see question 13.1.  If you're
	trying to convert an integer to a character, see question 8.6.
	If you're trying to set a pointer to point to a particular
	memory address, see question 19.25.


Section 5. Null Pointers

5.1:	What is this infamous null pointer, anyway?

A:	The language definition states that for each pointer type, there
	is a special value -- the "null pointer" -- which is
	distinguishable from all other pointer values and which is
	"guaranteed to compare unequal to a pointer to any object or
	function."  That is, the address-of operator & will never yield
	a null pointer, nor will a successful call to malloc().
	(malloc() does return a null pointer when it fails, and this is
	a typical use of null pointers: as a "special" pointer value
	with some other meaning, usually "not allocated" or "not
	pointing anywhere yet.")

	A null pointer is conceptually different from an uninitialized
	pointer.  A null pointer is known not to point to any object or
	function; an uninitialized pointer might point anywhere.  See
	also questions 1.30, 7.1, and 7.31.

	As mentioned above, there is a null pointer for each pointer
	type, and the internal values of null pointers for different
	types may be different.  Although programmers need not know the
	internal values, the compiler must always be informed which type
	of null pointer is required, so that it can make the distinction
	if necessary (see questions 5.2, 5.5, and 5.6 below).

	References: K&R1 Sec. 5.4 pp. 97-8; K&R2 Sec. 5.4 p. 102; ISO
	Sec. 6.2.2.3; Rationale Sec. 3.2.2.3; H&S Sec. 5.3.2 pp. 121-3.

5.2:	How do I get a null pointer in my programs?

A:	According to the language definition, a constant 0 in a pointer
	context is converted into a null pointer at compile time.  That
	is, in an initialization, assignment, or comparison when one
	side is a variable or expression of pointer type, the compiler
	can tell that a constant 0 on the other side requests a null
	pointer, and generate the correctly-typed null pointer value.
	Therefore, the following fragments are perfectly legal:

		char *p = 0;
		if(p != 0)

	(See also question 5.3.)

	However, an argument being passed to a function is not
	necessarily recognizable as a pointer context, and the compiler
	may not be able to tell that an unadorned 0 "means" a null
	pointer.  To generate a null pointer in a function call context,
	an explicit cast may be required, to force the 0 to be
	recognized as a pointer.  For example, the Unix system call
	execl takes a variable-length, null-pointer-terminated list of
	character pointer arguments, and is correctly called like this:

		execl("/bin/sh", "sh", "-c", "date", (char *)0);

	If the (char *) cast on the last argument were omitted, the
	compiler would not know to pass a null pointer, and would pass
	an integer 0 instead.  (Note that many Unix manuals get this
	example wrong.)

	When function prototypes are in scope, argument passing becomes
	an "assignment context," and most casts may safely be omitted,
	since the prototype tells the compiler that a pointer is
	required, and of which type, enabling it to correctly convert an
	unadorned 0.  Function prototypes cannot provide the types for
	variable arguments in variable-length argument lists however, so
	explicit casts are still required for those arguments.  (See
	also question 15.3.)  It is probably safest to properly cast
	all null pointer constants in function calls, to guard against
	varargs functions or those without prototypes.

	Summary:

		Unadorned 0 okay:	Explicit cast required:

		initialization		function call,
					no prototype in scope
		assignment
					variable argument in
		comparison		varargs function call

		function call,
		prototype in scope,
		fixed argument

	References: K&R1 Sec. A7.7 p. 190, Sec. A7.14 p. 192; K&R2
	Sec. A7.10 p. 207, Sec. A7.17 p. 209; ISO Sec. 6.2.2.3; H&S
	Sec. 4.6.3 p. 95, Sec. 6.2.7 p. 171.

5.3:	Is the abbreviated pointer comparison "if(p)" to test for non-
	null pointers valid?  What if the internal representation for
	null pointers is nonzero?

A:	When C requires the Boolean value of an expression, a false
	value is inferred when the expression compares equal to zero,
	and a true value otherwise.  That is, whenever one writes

		if(expr)

	where "expr" is any expression at all, the compiler essentially
	acts as if it had been written as

		if((expr) != 0)

	Substituting the trivial pointer expression "p" for "expr", we
	have

		if(p)	is equivalent to		if(p != 0)

	and this is a comparison context, so the compiler can tell that
	the (implicit) 0 is actually a null pointer constant, and use
	the correct null pointer value.  There is no trickery involved
	here; compilers do work this way, and generate identical code
	for both constructs.  The internal representation of a null
	pointer does *not* matter.

	The boolean negation operator, !, can be described as follows:

		!expr	is essentially equivalent to	(expr)?0:1
			or to				((expr) == 0)

	which leads to the conclusion that

		if(!p)	is equivalent to		if(p == 0)

	"Abbreviations" such as if(p), though perfectly legal, are
	considered by some to be bad style (and by others to be good
	style; see question 17.10).

	See also question 9.2.

	References: K&R2 Sec. A7.4.7 p. 204; ISO Sec. 6.3.3.3,
	Sec. 6.3.9, Sec. 6.3.13, Sec. 6.3.14, Sec. 6.3.15, Sec. 6.6.4.1,
	Sec. 6.6.5; H&S Sec. 5.3.2 p. 122.

5.4:	What is NULL and how is it defined?

A:	As a matter of style, many programmers prefer not to have
	unadorned 0's scattered through their programs.  Therefore, the
	preprocessor macro NULL is defined (by  and several
	other headers) as a null pointer constant, typically 0 or
	((void *)0) (see also question 5.6).  A programmer who wishes to
	make explicit the distinction between 0 the integer and 0 the
	null pointer constant can then use NULL whenever a null pointer
	is required.

	Using NULL is a stylistic convention only; the preprocessor
	turns NULL back into 0 which is then recognized by the compiler,
	in pointer contexts, as before.  In particular, a cast may still
	be necessary before NULL (as before 0) in a function call
	argument.  The table under question 5.2 above applies for NULL
	as well as 0 (an unadorned NULL is equivalent to an unadorned
	0).

	NULL should be used *only* as a pointer constant; see question 5.9.

	References: K&R1 Sec. 5.4 pp. 97-8; K&R2 Sec. 5.4 p. 102; ISO
	Sec. 7.1.6, Sec. 6.2.2.3; Rationale Sec. 4.1.5; H&S Sec. 5.3.2
	p. 122, Sec. 11.1 p. 292.

5.5:	How should NULL be defined on a machine which uses a nonzero bit
	pattern as the internal representation of a null pointer?

A:	The same as on any other machine: as 0 (or some version of 0;
	see question 5.4).

	Whenever a programmer requests a null pointer, either by writing
	"0" or "NULL", it is the compiler's responsibility to generate
	whatever bit pattern the machine uses for that null pointer.
	Therefore, #defining NULL as 0 on a machine for which internal
	null pointers are nonzero is as valid as on any other: the
	compiler must always be able to generate the machine's correct
	null pointers in response to unadorned 0's seen in pointer
	contexts.  See also questions 5.2, 5.10, and 5.17.

	References: ISO Sec. 7.1.6; Rationale Sec. 4.1.5.

5.6:	If NULL were defined as follows:

		#define NULL ((char *)0)

	wouldn't that make function calls which pass an uncast NULL
	work?

A:	Not in the most general case.  The complication is that there
	are machines which use different internal representations for
	pointers to different types of data.  The suggested definition
	would make uncast NULL arguments to functions expecting pointers
	to characters work correctly, but pointer arguments of other
	types could still (in the absence of prototypes) be
	problematical, and legal constructions such as

		FILE *fp = NULL;

	could fail.

	Nevertheless, ANSI C allows the alternate definition

		#define NULL ((void *)0)

	for NULL.  Besides potentially helping incorrect programs to
	work (but only on machines with homogeneous pointers, thus
	questionably valid assistance), this definition may catch
	programs which use NULL incorrectly (e.g. when the ASCII NUL
	character was really intended; see question 5.9).

	At any rate, ANSI function prototypes ensure that most (though
	not quite all; see question 5.2) pointer arguments are converted
	correctly when passed as function arguments, so the question is
	largely moot.

	References: Rationale Sec. 4.1.5.

5.9:	If NULL and 0 are equivalent as null pointer constants, which
	should I use?

A:	Many programmers believe that NULL should be used in all pointer
	contexts, as a reminder that the value is to be thought of as a
	pointer.  Others feel that the confusion surrounding NULL and 0
	is only compounded by hiding 0 behind a macro, and prefer to use
	unadorned 0 instead.  There is no one right answer.  (See also
	questions 9.2 and 17.10.)  C programmers must understand that
	NULL and 0 are interchangeable in pointer contexts, and that an
	uncast 0 is perfectly acceptable.  Any usage of NULL (as opposed
	to 0) should be considered a gentle reminder that a pointer is
	involved; programmers should not depend on it (either for their
	own understanding or the compiler's) for distinguishing pointer
	0's from integer 0's.

	NULL should *not* be used when another kind of 0 is required,
	even though it might work, because doing so sends the wrong
	stylistic message.  (Furthermore, ANSI allows the definition of
	NULL to be ((void *)0), which will not work at all in non-
	pointer contexts.)  In particular, do not use NULL when the
	ASCII null character (NUL) is desired.  Provide your own
	definition

		#define NUL '/0'

	if you must.

	References: K&R1 Sec. 5.4 pp. 97-8; K&R2 Sec. 5.4 p. 102.

5.10:	But wouldn't it be better to use NULL (rather than 0), in case
	the value of NULL changes, perhaps on a machine with nonzero
	internal null pointers?

A:	No.  (Using NULL may be preferable, but not for this reason.)
	Although symbolic constants are often used in place of numbers
	because the numbers might change, this is *not* the reason that
	NULL is used in place of 0.  Once again, the language guarantees
	that source-code 0's (in pointer contexts) generate null
	pointers.  NULL is used only as a stylistic convention.  See
	questions 5.5 and 9.2.

5.12:	I use the preprocessor macro

		#define Nullptr(type) (type *)0

	to help me build null pointers of the correct type.

A:	This trick, though popular and superficially attractive, does
	not buy much.  It is not needed in assignments or comparisons;
	see question 5.2.  (It does not even save keystrokes.)  See also
	questions 9.1 and 10.2.

5.13:	This is strange.  NULL is guaranteed to be 0, but the null
	pointer is not?

A:	When the term "null" or "NULL" is casually used, one of several
	things may be meant:

	1.	The conceptual null pointer, the abstract language concept
		defined in question 5.1.  It is implemented with...

	2.	The internal (or run-time) representation of a null
		pointer, which may or may not be all-bits-0 and which may
		be different for different pointer types.  The actual
		values should be of concern only to compiler writers.
		Authors of C programs never see them, since they use...

	3.	The null pointer constant, which is a constant integer 0
		(see question 5.2).  It is often hidden behind...

	4.	The NULL macro, which is #defined to be 0 (see question
		5.4).  Finally, as red herrings, we have...

	5.	The ASCII null character (NUL), which does have all bits
		zero, but has no necessary relation to the null pointer
		except in name; and...

	6.	The "null string," which is another name for the empty
		string ("").  Using the term "null string" can be
		confusing in C, because an empty string involves a null
		('/0') character, but *not* a null pointer, which brings
		us full circle...

	This article uses the phrase "null pointer" (in lower case) for
	sense 1, the token "0" or the phrase "null pointer constant"
	for sense 3, and the capitalized word "NULL" for sense 4.

5.14:	Why is there so much confusion surrounding null pointers?  Why
	do these questions come up so often?

A:	C programmers traditionally like to know a lot (perhaps more
	than they need to) about the underlying machine implementation.
	The fact that null pointers are represented both in source code,
	and internally to most machines, as zero invites unwarranted
	assumptions.  The use of a preprocessor macro (NULL) may seem to
	suggest that the value could change some day, or on some weird
	machine.  The construct "if(p == 0)" is easily misread as
	calling for conversion of p to an integral type, rather than
	0 to a pointer type, before the comparison.  Finally, the
	distinction between the several uses of the term "null"
	(listed in question 5.13 above) is often overlooked.

	One good way to wade out of the confusion is to imagine that C
	used a keyword (perhaps "nil", like Pascal) as a null pointer
	constant.  The compiler could either turn "nil" into the
	appropriate type of null pointer when it could unambiguously
	determine that type from the source code, or complain when it
	could not.  Now in fact, in C the keyword for a null pointer
	constant is not "nil" but "0", which works almost as well,
	except that an uncast "0" in a non-pointer context generates an
	integer zero instead of an error message, and if that uncast 0
	was supposed to be a null pointer constant, the resulting
	program may not work.

5.15:	I'm confused.  I just can't understand all this null pointer
	stuff.

A:	Here are two simple rules you can follow:

	1.	When you want a null pointer constant in source code,
		use "0" or "NULL".

	2.	If the usage of "0" or "NULL" is an argument in a
		function call, cast it to the pointer type expected by
		the function being called.

	The rest of the discussion has to do with other people's
	misunderstandings, with the internal representation of null
	pointers (which you shouldn't need to know), and with the
	complexities of function prototypes.  (Taking those complexities
	into account, we find that rule 2 is conservative, of course;
	but it doesn't hurt.)  Understand questions 5.1, 5.2, and 5.4,
	and consider 5.3, 5.9, 5.13, and 5.14, and you'll do fine.

5.16:	Given all the confusion surrounding null pointers, wouldn't it
	be easier simply to require them to be represented internally by
	zeroes?

A:	If for no other reason, doing so would be ill-advised because it
	would unnecessarily constrain implementations which would
	otherwise naturally represent null pointers by special, nonzero
	bit patterns, particularly when those values would trigger
	automatic hardware traps for invalid accesses.

	Besides, what would such a requirement really accomplish?
	Proper understanding of null pointers does not require knowledge
	of the internal representation, whether zero or nonzero.
	Assuming that null pointers are internally zero does not make
	any code easier to write (except for a certain ill-advised usage
	of calloc(); see question 7.31).  Known-zero internal pointers
	would not obviate casts in function calls, because the *size* of
	the pointer might still be different from that of an int.  (If
	"nil" were used to request null pointers, as mentioned in
	question 5.14 above, the urge to assume an internal zero
	representation would not even arise.)

5.17:	Seriously, have any actual machines really used nonzero null
	pointers, or different representations for pointers to different
	types?

A:	The Prime 50 series used segment 07777, offset 0 for the null
	pointer, at least for PL/I.  Later models used segment 0, offset
	0 for null pointers in C, necessitating new instructions such as
	TCNP (Test C Null Pointer), evidently as a sop to all the extant
	poorly-written C code which made incorrect assumptions.  Older,
	word-addressed Prime machines were also notorious for requiring
	larger byte pointers (char *'s) than word pointers (int *'s).

	The Eclipse MV series from Data General has three
	architecturally supported pointer formats (word, byte, and bit
	pointers), two of which are used by C compilers: byte pointers
	for char * and void *, and word pointers for everything else.

	Some Honeywell-Bull mainframes use the bit pattern 06000 for
	(internal) null pointers.

	The CDC Cyber 180 Series has 48-bit pointers consisting of a
	ring, segment, and offset.  Most users (in ring 11) have null
	pointers of 0xB00000000000.  It was common on old CDC ones-
	complement machines to use an all-one-bits word as a special
	flag for all kinds of data, including invalid addresses.

	The old HP 3000 series uses a different addressing scheme for
	byte addresses than for word addresses; like several of the
	machines above it therefore uses different representations for
	char * and void * pointers than for other pointers.

	The Symbolics Lisp Machine, a tagged architecture, does not even
	have conventional numeric pointers; it uses the pair 
	(basically a nonexistent  handle) as a C null
	pointer.

	Depending on the "memory model" in use, 8086-family processors
	(PC compatibles) may use 16-bit data pointers and 32-bit
	function pointers, or vice versa.

	Some 64-bit Cray machines represent int * in the lower 48 bits
	of a word; char * additionally uses some of the upper 16 bits to
	indicate a byte address within a word.

	References: K&R1 Sec. A14.4 p. 211.

5.20:	What does a run-time "null pointer assignment" error mean?

A:	This message, which typically occurs with MS-DOS compilers,
	means that you've written, via a null pointer, to an invalid
	location -- probably offset 0 in the default data segment.
	See also question 16.8.


Section 6. Arrays and Pointers

6.1:	I had the definition char a[6] in one source file, and in
	another I declared extern char *a.  Why didn't it work?

A:	In one source file you defined an array of characters and in the
	other you declared a pointer to characters.  The declaration
	extern char *a simply does not match the actual definition.
	The type pointer-to-type-T is not the same as array-of-type-T.
	Use extern char a[].

	References: ISO Sec. 6.5.4.2; CT&P Sec. 3.3 pp. 33-4, Sec. 4.5
	pp. 64-5.

6.2:	But I heard that char a[] was identical to char *a.

A:	Not at all.  (What you heard has to do with formal parameters to
	functions; see question 6.4.)  Arrays are not pointers.  The
	array declaration char a[6] requests that space for six
	characters be set aside, to be known by the name "a".  That is,
	there is a location named "a" at which six characters can sit.
	The pointer declaration char *p, on the other hand, requests a
	place which holds a pointer, to be known by the name "p".  This
	pointer can point almost anywhere: to any char, or to any
	contiguous array of chars, or nowhere (see also questions 5.1
	and 1.30).

	As usual, a picture is worth a thousand words.  The declarations

		char a[] = "hello";
		char *p = "world";

	would initialize data structures which could be represented like
	this:
		   +---+---+---+---+---+---+
		a: | h | e | l | l | o |/0 |
		   +---+---+---+---+---+---+
		   +-----+     +---+---+---+---+---+---+
		p: |  *======> | w | o | r | l | d |/0 |
		   +-----+     +---+---+---+---+---+---+

	It is useful to realize that a reference like x[3] generates
	different code depending on whether x is an array or a pointer.
	Given the declarations above, when the compiler sees the
	expression a[3], it emits code to start at the location "a",
	move three past it, and fetch the character there.  When it sees
	the expression p[3], it emits code to start at the location "p",
	fetch the pointer value there, add three to the pointer, and
	finally fetch the character pointed to.  In other words, a[3] is
	three places past (the start of) the object *named* a, while
	p[3] is three places past the object *pointed to* by p.  In the
	example above, both a[3] and p[3] happen to be the character
	'l', but the compiler gets there differently.  (The essential
	difference is that the values of an array like a and a pointer
	like p are computed differently *whenever* they appear in
	expressions, whether or not they are being subscripted, as
	explained further in the next question.)  See also question 1.32.

	References: K&R2 Sec. 5.5 p. 104; CT&P Sec. 4.5 pp. 64-5.

6.3:	So what is meant by the "equivalence of pointers and arrays" in
	C?

A:	Much of the confusion surrounding arrays and pointers in C can
	be traced to a misunderstanding of this statement.  Saying that
	arrays and pointers are "equivalent" means neither that they are
	identical nor even interchangeable.  What it means is that array
	and pointer arithmetic is defined such that a pointer can be
	conveniently used to access an array or to simulate an array.

	Specifically, the cornerstone of the equivalence is this key
	definition:

		An lvalue of type array-of-T which appears in an
		expression decays (with three exceptions) into a
		pointer to its first element; the type of the
		resultant pointer is pointer-to-T.

	That is, whenever an array appears in an expression,
	the compiler implicitly generates a pointer to the array's
	first element, just as if the programmer had written &a[0].
	(The exceptions are when the array is the operand of a sizeof or
	& operator, or is a string literal initializer for a character
	array.)

	As a consequence of this definition, the compiler doesn't apply
	the array subscripting operator [] that differently to arrays
	and pointers, after all.  In an expression of the form a[i], the
	array decays into a pointer, following the rule above, and is
	then subscripted just as would be a pointer variable in the
	expression p[i] (although the eventual memory accesses will be
	different, as explained in question 6.2).  If you were to assign
	the array's address to the pointer:

		p = a;

	then p[3] and a[3] would access the same element.

	See also questions 6.8 and 6.14.

	References: K&R1 Sec. 5.3 pp. 93-6; K&R2 Sec. 5.3 p. 99; ISO
	Sec. 6.2.2.1, Sec. 6.3.2.1, Sec. 6.3.6; H&S Sec. 5.4.1 p. 124.

6.4:	Then why are array and pointer declarations interchangeable as
	function formal parameters?

A:	It's supposed to be a convenience.

	Since arrays decay immediately into pointers, an array is never
	actually passed to a function.  Allowing pointer parameters to
	be declared as arrays is a simply a way of making it look as
	though an array was being passed, perhaps because the parameter
	will be used within the function as if it were an array.
	Specifically, any parameter declarations which "look like"
	arrays, e.g.

		void f(char a[])
		{ ... }

	are treated by the compiler as if they were pointers, since that
	is what the function will receive if an array is passed:

		void f(char *a)
		{ ... }

	This conversion holds only within function formal parameter
	declarations, nowhere else.  If the conversion bothers you,
	avoid it; many programmers have concluded that the confusion it
	causes outweighs the small advantage of having the declaration
	"look like" the call or the uses within the function.

	See also question 6.21.

	References: K&R1 Sec. 5.3 p. 95, Sec. A10.1 p. 205; K&R2
	Sec. 5.3 p. 100, Sec. A8.6.3 p. 218, Sec. A10.1 p. 226; ISO
	Sec. 6.5.4.3, Sec. 6.7.1, Sec. 6.9.6; H&S Sec. 9.3 p. 271; CT&P
	Sec. 3.3 pp. 33-4.

6.7:	How can an array be an lvalue, if you can't assign to it?

A:	The ANSI C Standard defines a "modifiable lvalue," which an
	array is not.

	References: ISO Sec. 6.2.2.1; Rationale Sec. 3.2.2.1; H&S
	Sec. 7.1 p. 179.

6.8:	Practically speaking, what is the difference between arrays and
	pointers?

A:	Arrays automatically allocate space, but can't be relocated or
	resized.  Pointers must be explicitly assigned to point to
	allocated space (perhaps using malloc), but can be reassigned
	(i.e. pointed at different objects) at will, and have many other
	uses besides serving as the base of blocks of memory.

	Due to the so-called equivalence of arrays and pointers (see
	question 6.3), arrays and pointers often seem interchangeable,
	and in particular a pointer to a block of memory assigned by
	malloc is frequently treated (and can be referenced using [])
	exactly as if it were a true array.  See questions 6.14 and
	6.16.  (Be careful with sizeof, though.)

	See also questions 1.32 and 20.14.

6.9:	Someone explained to me that arrays were really just constant
	pointers.

A:	This is a bit of an oversimplification.  An array name is
	"constant" in that it cannot be assigned to, but an array is
	*not* a pointer, as the discussion and pictures in question 6.2
	should make clear.  See also questions 6.3 and 6.8.

6.11:	I came across some "joke" code containing the "expression"
	5["abcdef"] .  How can this be legal C?

A:	Yes, Virginia, array subscripting is commutative in C.  This
	curious fact follows from the pointer definition of array
	subscripting, namely that a[e] is identical to *((a)+(e)), for
	*any* two expressions a and e, as long as one of them is a
	pointer expression and one is integral.  This unsuspected
	commutativity is often mentioned in C texts as if it were
	something to be proud of, but it finds no useful application
	outside of the Obfuscated C Contest (see question 20.36).

	References: Rationale Sec. 3.3.2.1; H&S Sec. 5.4.1 p. 124,
	Sec. 7.4.1 pp. 186-7.

6.12:	Since array references decay into pointers, if arr is an array,
	what's the difference between arr and &arr?

A:	The type.

	In Standard C, &arr yields a pointer, of type pointer-to-array-
	of-T, to the entire array.  (In pre-ANSI C, the & in &arr
	generally elicited a warning, and was generally ignored.)  Under
	all C compilers, a simple reference (without an explicit &) to
	an array yields a pointer, of type pointer-to-T, to the array's
	first element.  (See also questions 6.3, 6.13, and 6.18.)

	References: ISO Sec. 6.2.2.1, Sec. 6.3.3.2; Rationale
	Sec. 3.3.3.2; H&S Sec. 7.5.6 p. 198.

6.13:	How do I declare a pointer to an array?

A:	Usually, you don't want to.  When people speak casually of a
	pointer to an array, they usually mean a pointer to its first
	element.

	Instead of a pointer to an array, consider using a pointer to
	one of the array's elements.  Arrays of type T decay into
	pointers to type T (see question 6.3), which is convenient;
	subscripting or incrementing the resultant pointer will access
	the individual members of the array.  True pointers to arrays,
	when subscripted or incremented, step over entire arrays, and
	are generally useful only when operating on arrays of arrays, if
	at all.  (See question 6.18.)

	If you really need to declare a pointer to an entire array, use
	something like "int (*ap)[N];" where N is the size of the array.
	(See also question 1.21.)  If the size of the array is unknown,
	N can in principle be omitted, but the resulting type, "pointer
	to array of unknown size," is useless.

	See also question 6.12 above.

	References: ISO Sec. 6.2.2.1.

6.14:	How can I set an array's size at run time?
	How can I avoid fixed-sized arrays?

A:	The equivalence between arrays and pointers (see question 6.3)
	allows a pointer to malloc'ed memory to simulate an array
	quite effectively.  After executing

		#include 
		int *dynarray;
		dynarray = malloc(10 * sizeof(int));

	(and if the call to malloc succeeds), you can reference
	dynarray[i] (for i from 0 to 9) almost as if dynarray were a
	conventional, statically-allocated array (int a[10]).  The only
	difference is that sizeof will not give the size of the "array".
	See also questions 1.31b, 6.16, and 7.7.

6.15:	How can I declare local arrays of a size matching a passed-in
	array?

A:	Until recently, you couldn't; array dimensions in C
	traditionally had to be compile-time constants.  However, C99
	introduces variable-length arrays (VLA's) which solve this
	problem; local arrays may have sizes set by variables or other
	expressions, perhaps involving function parameters.  (gcc has
	provided parameterized arrays as an extension for some time.)
	If you can't use C99 or gcc, you'll have to use malloc(), and
	remember to call free() before the function returns.  See also
	questions 6.14, 6.16, 6.19, 7.22, and maybe 7.32.

	References: ISO Sec. 6.4, Sec. 6.5.4.2; C9X Sec. 6.5.5.2.

6.16:	How can I dynamically allocate a multidimensional array?

A:	The traditional solution is to allocate an array of pointers,
	and then initialize each pointer to a dynamically-allocated
	"row."  Here is a two-dimensional example:

		#include 

		int **array1 = malloc(nrows * sizeof(int *));
		for(i = 0; i < nrows; i++)
			array1[i] = malloc(ncolumns * sizeof(int));

	(In real code, of course, all of malloc's return values would be
	checked.  You can also use sizeof(*array1) and sizeof(**array1)
	instead of sizeof(int *) and sizeof(int).)

	You can keep the array's contents contiguous, at the cost of
	making later reallocation of individual rows more difficult,
	with a bit of explicit pointer arithmetic:

		int **array2 = malloc(nrows * sizeof(int *));
		array2[0] = malloc(nrows * ncolumns * sizeof(int));
		for(i = 1; i < nrows; i++)
			array2[i] = array2[0] + i * ncolumns;

	In either case, the elements of the dynamic array can be
	accessed with normal-looking array subscripts: arrayx[i][j]
	(for 0 <= i < nrows and 0 <= j < ncolumns).

	If the double indirection implied by the above schemes is for
	some reason unacceptable, you can simulate a two-dimensional
	array with a single, dynamically-allocated one-dimensional
	array:

		int *array3 = malloc(nrows * ncolumns * sizeof(int));

	However, you must now perform subscript calculations manually,
	accessing the i,jth element with array3[i * ncolumns + j].  (A
	macro could hide the explicit calculation, but invoking it would
	require parentheses and commas which wouldn't look exactly like
	multidimensional array syntax, and the macro would need access
	to at least one of the dimensions, as well.  See also question
	6.19.)

	Yet another option is to use pointers to arrays:

		int (*array4)[NCOLUMNS] = malloc(nrows * sizeof(*array4));

	but the syntax starts getting horrific and at most one dimension
	may be specified at run time.

	With all of these techniques, you may of course need to remember
	to free the arrays (which may take several steps; see question
	7.23) when they are no longer needed, and you cannot necessarily
	intermix dynamically-allocated arrays with conventional,
	statically-allocated ones (see question 6.20, and also question
	6.18).

	Finally, in C99 you can use a variable-length array.

	All of these techniques can also be extended to three or more
	dimensions.

	References: C9X Sec. 6.5.5.2.

6.17:	Here's a neat trick: if I write

		int realarray[10];
		int *array = &realarray[-1];

	I can treat "array" as if it were a 1-based array.

A:	Although this technique is attractive (and was used in old
	editions of the book _Numerical Recipes in C_), it is not
	strictly conforming to the C Standard.  Pointer arithmetic
	is defined only as long as the pointer points within the same
	allocated block of memory, or to the imaginary "terminating"
	element one past it; otherwise, the behavior is undefined,
	*even if the pointer is not dereferenced*.  The code above
	could fail if, while subtracting the offset, an illegal
	address were generated (perhaps because the address tried
	to "wrap around" past the beginning of some memory segment).

	References: K&R2 Sec. 5.3 p. 100, Sec. 5.4 pp. 102-3, Sec. A7.7
	pp. 205-6; ISO Sec. 6.3.6; Rationale Sec. 3.2.2.3.

6.18:	My compiler complained when I passed a two-dimensional array to
	a function expecting a pointer to a pointer.

A:	The rule (see question 6.3) by which arrays decay into pointers
	is *not* applied recursively.  An array of arrays (i.e. a two-
	dimensional array in C) decays into a pointer to an array, not a
	pointer to a pointer.  Pointers to arrays can be confusing, and
	must be treated carefully; see also question 6.13.

	If you are passing a two-dimensional array to a function:

		int array[NROWS][NCOLUMNS];
		f(array);

	the function's declaration must match:

		void f(int a[][NCOLUMNS])
		{ ... }

	or

		void f(int (*ap)[NCOLUMNS])	/* ap is a pointer to an array */
		{ ... }

	In the first declaration, the compiler performs the usual
	implicit parameter rewriting of "array of array" to "pointer to
	array" (see questions 6.3 and 6.4); in the second form the
	pointer declaration is explicit.  Since the called function does
	not allocate space for the array, it does not need to know the
	overall size, so the number of rows, NROWS, can be omitted.  The
	width of the array is still important, so the column dimension
	NCOLUMNS (and, for three- or more dimensional arrays, the
	intervening ones) must be retained.

	If a function is already declared as accepting a pointer to a
	pointer, it is almost certainly meaningless to pass a two-
	dimensional array directly to it.

	See also questions 6.12 and 6.15.

	References: K&R1 Sec. 5.10 p. 110; K&R2 Sec. 5.9 p. 113; H&S
	Sec. 5.4.3 p. 126.

6.19:	How do I write functions which accept two-dimensional arrays
	when the width is not known at compile time?

A:	It's not always easy.  One way is to pass in a pointer to the
	[0][0] element, along with the two dimensions, and simulate
	array subscripting "by hand":

		void f2(int *aryp, int nrows, int ncolumns)
		{ ... array[i][j] is accessed as aryp[i * ncolumns + j] ... }

	This function could be called with the array from question 6.18
	as

		f2(&array[0][0], NROWS, NCOLUMNS);

	It must be noted, however, that a program which performs
	multidimensional array subscripting "by hand" in this way is not
	in strict conformance with the ANSI C Standard; according to an
	official interpretation, the behavior of accessing
	(&array[0][0])[x] is not defined for x >= NCOLUMNS.

	C99 allows variable-length arrays, and once compilers which
	accept C99's extensions become widespread, VLA's will probably
	become the preferred solution.  (gcc has supported variable-
	sized arrays for some time.)

	When you want to be able to use a function on multidimensional
	arrays of various sizes, one solution is to simulate all the
	arrays dynamically, as in question 6.16.

	See also questions 6.18, 6.20, and 6.15.

	References: ISO Sec. 6.3.6; C9X Sec. 6.5.5.2.

6.20:	How can I use statically- and dynamically-allocated
	multidimensional arrays interchangeably when passing them to
	functions?

A:	There is no single perfect method.  Given the declarations

		int array[NROWS][NCOLUMNS];
		int **array1;			/* ragged */
		int **array2;			/* contiguous */
		int *array3;			/* "flattened" */
		int (*array4)[NCOLUMNS];

	with the pointers initialized as in the code fragments in
	question 6.16, and functions declared as

		void f1a(int a[][NCOLUMNS], int nrows, int ncolumns);
		void f1b(int (*a)[NCOLUMNS], int nrows, int ncolumns);
		void f2(int *aryp, int nrows, int ncolumns);
		void f3(int **pp, int nrows, int ncolumns);

	where f1a() and f1b() accept conventional two-dimensional
	arrays, f2() accepts a "flattened" two-dimensional array, and
	f3() accepts a pointer-to-pointer, simulated array (see also
	questions 6.18 and 6.19), the following calls should work as
	expected:

		f1a(array, NROWS, NCOLUMNS);
		f1b(array, NROWS, NCOLUMNS);
		f1a(array4, nrows, NCOLUMNS);
		f1b(array4, nrows, NCOLUMNS);
		f2(&array[0][0], NROWS, NCOLUMNS);
		f2(*array, NROWS, NCOLUMNS);
		f2(*array2, nrows, ncolumns);
		f2(array3, nrows, ncolumns);
		f2(*array4, nrows, NCOLUMNS);
		f3(array1, nrows, ncolumns);
		f3(array2, nrows, ncolumns);

	The following calls would probably work on most systems, but
	involve questionable casts, and work only if the dynamic
	ncolumns matches the static NCOLUMNS:

		f1a((int (*)[NCOLUMNS])(*array2), nrows, ncolumns);
		f1a((int (*)[NCOLUMNS])(*array2), nrows, ncolumns);
		f1b((int (*)[NCOLUMNS])array3, nrows, ncolumns);
		f1b((int (*)[NCOLUMNS])array3, nrows, ncolumns);

	It must again be noted that passing &array[0][0] (or,
	equivalently, *array) to f2() is not strictly conforming; see
	question 6.19.

	If you can understand why all of the above calls work and are
	written as they are, and if you understand why the combinations
	that are not listed would not work, then you have a *very* good
	understanding of arrays and pointers in C.

	Rather than worrying about all of this, one approach to using
	multidimensional arrays of various sizes is to make them *all*
	dynamic, as in question 6.16.  If there are no static
	multidimensional arrays -- if all arrays are allocated like
	array1 or array2 in question 6.16 -- then all functions can be
	written like f3().

6.21:	Why doesn't sizeof properly report the size of an array when the
	array is a parameter to a function?

A:	The compiler pretends that the array parameter was declared as a
	pointer (see question 6.4), and sizeof reports the size of the
	pointer.

	References: H&S Sec. 7.5.2 p. 195.


Section 7. Memory Allocation

7.1:	Why doesn't this fragment work?

		char *answer;
		printf("Type something:/n");
		gets(answer);
		printf("You typed /"%s/"/n", answer);

A:	The pointer variable answer, which is handed to gets() as the
	location into which the response should be stored, has not been
	set to point to any valid storage.  That is, we cannot say where
	the pointer answer points.  (Since local variables are not
	initialized, and typically contain garbage, it is not even
	guaranteed that answer starts out as a null pointer.
	See questions 1.30 and 5.1.)

	The simplest way to correct the question-asking program is to
	use a local array, instead of a pointer, and let the compiler
	worry about allocation:

		#include 
		#include 

		char answer[100], *p;
		printf("Type something:/n");
		fgets(answer, sizeof answer, stdin);
		if((p = strchr(answer, '/n')) != NULL)
			*p = '/0';
		printf("You typed /"%s/"/n", answer);

	This example also uses fgets() instead of gets(), so that the
	end of the array cannot be overwritten.  (See question 12.23.
	Unfortunately for this example, fgets() does not automatically
	delete the trailing /n, as gets() would.)  It would also be
	possible to use malloc() to allocate the answer buffer.

7.2:	I can't get strcat() to work.  I tried

		char *s1 = "Hello, ";
		char *s2 = "world!";
		char *s3 = strcat(s1, s2);

	but I got strange results.

A:	As in question 7.1 above, the main problem here is that space
	for the concatenated result is not properly allocated.  C does
	not provide an automatically-managed string type.  C compilers
	allocate memory only for objects explicitly mentioned in the
	source code (in the case of strings, this includes character
	arrays and string literals).  The programmer must arrange for
	sufficient space for the results of run-time operations such as
	string concatenation, typically by declaring arrays, or by
	calling malloc().

	strcat() performs no allocation; the second string is appended
	to the first one, in place.  Therefore, one fix would be to
	declare the first string as an array:

		char s1[20] = "Hello, ";

	Since strcat() returns the value of its first argument (s1, in
	this case), the variable s3 is superfluous; after the call to
	strcat(), s1 contains the result.

	The original call to strcat() in the question actually has two
	problems: the string literal pointed to by s1, besides not being
	big enough for any concatenated text, is not necessarily
	writable at all.  See question 1.32.

	References: CT&P Sec. 3.2 p. 32.

7.3:	But the man page for strcat() says that it takes two char *'s as
	arguments.  How am I supposed to know to allocate things?

A:	In general, when using pointers you *always* have to consider
	memory allocation, if only to make sure that the compiler is
	doing it for you.  If a library function's documentation does
	not explicitly mention allocation, it is usually the caller's
	problem.

	The Synopsis section at the top of a Unix-style man page or in
	the ANSI C standard can be misleading.  The code fragments
	presented there are closer to the function definitions used by
	an implementor than the invocations used by the caller.  In
	particular, many functions which accept pointers (e.g. to
	structures or strings) are usually called with a pointer to some
	object (a structure, or an array -- see questions 6.3 and 6.4)
	which the caller has allocated.  Other common examples are
	time() (see question 13.12) and stat().

7.3b:	I just tried the code

		char *p;
		strcpy(p, "abc");

	and it worked.  How?  Why didn't it crash?

A:	You got lucky, I guess.  The memory randomly pointed to by
	the uninitialized pointer p happened to be writable by you,
	and apparently was not already in use for anything vital.
	See also question 11.35.

7.3c:	How much memory does a pointer variable allocate?

A:	That's a pretty misleading question.  When you declare
	a pointer variable, as in

		char *p;

	you (or, more properly, the compiler) have allocated only enough
	memory to hold the pointer itself; that is, in this case you
	have allocated sizeof(char *) bytes of memory.  But you have
	not yet allocated *any* memory for the pointer to point to.
	See also questions 7.1 and 7.2.

7.5a:	I have a function that is supposed to return a string, but when
	it returns to its caller, the returned string is garbage.

A:	Make sure that the pointed-to memory is properly allocated.
	For example, make sure you have *not* done something like

		char *itoa(int n)
		{
			char retbuf[20];		/* WRONG */
			sprintf(retbuf, "%d", n);
			return retbuf;			/* WRONG */
		}

	One fix (which is imperfect, especially if the function in
	question is called recursively, or if several of its return
	values are needed simultaneously) would be to declare the return
	buffer as

			static char retbuf[20];

	See also questions 7.5b, 12.21, and 20.1.

	References: ISO Sec. 6.1.2.4.

7.5b:	So what's the right way to return a string or other aggregate?

A:	The returned pointer should be to a statically-allocated buffer
	(as in the answer to question 7.5a), or to a buffer passed in by
	the caller, or to memory obtained with malloc(), but *not* to a
	local (automatic) array.

	See also question 20.1.

7.6:	Why am I getting "warning: assignment of pointer from integer
	lacks a cast" for calls to malloc()?

A:	Have you #included , or otherwise arranged for
	malloc() to be declared properly?  See also question 1.25.

	References: H&S Sec. 4.7 p. 101.

7.7:	Why does some code carefully cast the values returned by malloc
	to the pointer type being allocated?

A:	Before ANSI/ISO Standard C introduced the void * generic pointer
	type, these casts were typically required to silence warnings
	(and perhaps induce conversions) when assigning between
	incompatible pointer types.

	Under ANSI/ISO Standard C, these casts are no longer necessary,
	and in fact modern practice discourages them, since they can
	camouflage important warnings which would otherwise be generated
	if malloc() happened not to be declared correctly; see question
	7.6 above.  (However, the casts are typically seen in C code
	which for one reason or another is intended to be compatible
	with C++, where explicit casts from void * are required.)

	References: H&S Sec. 16.1 pp. 386-7.

7.7c:	In a call to malloc(), what does an error like "Cannot convert
	`void *' to `int *'" mean?

A:	It means you're using a C++ compiler instead of a C compiler.
	See question 7.7.

7.8:	I see code like

		char *p = malloc(strlen(s) + 1);
		strcpy(p, s);

	Shouldn't that be malloc((strlen(s) + 1) * sizeof(char))?

A:	It's never necessary to multiply by sizeof(char), since
	sizeof(char) is, by definition, exactly 1.  (On the other
	hand, multiplying by sizeof(char) doesn't hurt, and in some
	circumstances may help by introducing a size_t into the
	expression.)  See also question 8.9.

	References: ISO Sec. 6.3.3.4; H&S Sec. 7.5.2 p. 195.

7.11:	How can I dynamically allocate arrays?

A:	See questions 6.14 and 6.16.

7.14:	I've heard that some operating systems don't actually allocate
	malloc'ed memory until the program tries to use it.  Is this
	legal?

A:	It's hard to say.  The Standard doesn't say that systems can act
	this way, but it doesn't explicitly say that they can't, either.

	References: ISO Sec. 7.10.3.

7.16:	I'm allocating a large array for some numeric work, using the
	line

		double *array = malloc(300 * 300 * sizeof(double));

	malloc() isn't returning null, but the program is acting
	strangely, as if it's overwriting memory, or malloc() isn't
	allocating as much as I asked for, or something.

A:	Notice that 300 x 300 is 90,000, which will not fit in a 16-bit
	int, even before you multiply it by sizeof(double).  If you
	need to allocate this much memory, you'll have to be careful.
	If size_t (the type accepted by malloc()) is a 32-bit type on
	your machine, but int is 16 bits, you might be able to get away
	with writing 300 * (300 * sizeof(double)) (see question 3.14).
	Otherwise, you'll have to break your data structure up into
	smaller chunks, or use a 32-bit machine or compiler, or use
	some nonstandard memory allocation functions.  See also
	question 19.23.

7.17:	I've got 8 meg of memory in my PC.  Why can I only seem to
	malloc 640K or so?

A:	Under the segmented architecture of PC compatibles, it can be
	difficult to use more than 640K with any degree of transparency,
	especially under MS-DOS.  See also question 19.23.

7.19:	My program is crashing, apparently somewhere down inside malloc,
	but I can't see anything wrong with it.  Is there a bug in
	malloc()?

A:	It is unfortunately very easy to corrupt malloc's internal data
	structures, and the resulting problems can be stubborn.  The
	most common source of problems is writing more to a malloc'ed
	region than it was allocated to hold; a particularly common bug
	is to malloc(strlen(s)) instead of strlen(s) + 1.  Other
	problems may involve using pointers to memory that has been
	freed, freeing pointers twice, freeing pointers not obtained
	from malloc, or trying to realloc a null pointer (see question
	7.30).

	See also questions 7.26, 16.8, and 18.2.

7.20:	You can't use dynamically-allocated memory after you free it,
	can you?

A:	No.  Some early documentation for malloc() stated that the
	contents of freed memory were "left undisturbed," but this ill-
	advised guarantee was never universal and is not required by the
	C Standard.

	Few programmers would use the contents of freed memory
	deliberately, but it is easy to do so accidentally.  Consider
	the following (correct) code for freeing a singly-linked list:

		struct list *listp, *nextp;
		for(listp = base; listp != NULL; listp = nextp) {
			nextp = listp->next;
			free(listp);
		}

	and notice what would happen if the more-obvious loop iteration
	expression listp = listp->next were used, without the temporary
	nextp pointer.

	References: K&R2 Sec. 7.8.5 p. 167; ISO Sec. 7.10.3; Rationale
	Sec. 4.10.3.2; H&S Sec. 16.2 p. 387; CT&P Sec. 7.10 p. 95.

7.21:	Why isn't a pointer null after calling free()?
	How unsafe is it to use (assign, compare) a pointer value after
	it's been freed?

A:	When you call free(), the memory pointed to by the passed
	pointer is freed, but the value of the pointer in the caller
	probably remains unchanged, because C's pass-by-value semantics
	mean that called functions never permanently change the values
	of their arguments.  (See also question 4.8.)

	A pointer value which has been freed is, strictly speaking,
	invalid, and *any* use of it, even if it is not dereferenced,
	can theoretically lead to trouble, though as a quality of
	implementation issue, most implementations will probably not go
	out of their way to generate exceptions for innocuous uses of
	invalid pointers.

	References: ISO Sec. 7.10.3; Rationale Sec. 3.2.2.3.

7.22:	When I call malloc() to allocate memory for a pointer which is
	local to a function, do I have to explicitly free() it?

A:	Yes.  Remember that a pointer is different from what it points
	to.  Local variables are deallocated when the function returns,
	but in the case of a pointer variable, this means that the
	pointer is deallocated, *not* what it points to.  Memory
	allocated with malloc() always persists until you explicitly
	free it.  In general, for every call to malloc(), there should
	be a corresponding call to free().

7.23:	I'm allocating structures which contain pointers to other
	dynamically-allocated objects.  When I free a structure, do I
	also have to free each subsidiary pointer?

A:	Yes.  In general, you must arrange that each pointer returned
	from malloc() be individually passed to free(), exactly once (if
	it is freed at all).  A good rule of thumb is that for each call
	to malloc() in a program, you should be able to point at the
	call to free() which frees the memory allocated by that malloc()
	call.

	See also question 7.24.

7.24:	Must I free allocated memory before the program exits?

A:	You shouldn't have to.  A real operating system definitively
	reclaims all memory and other resources when a program exits.
	Nevertheless, some personal computers are said not to reliably
	recover memory, and all that can be inferred from the ANSI/ISO C
	Standard is that this is a "quality of implementation issue."

	References: ISO Sec. 7.10.3.2.

7.25:	I have a program which mallocs and later frees a lot of memory,
	but I can see from the operating system that memory usage
	doesn't actually go back down.

A:	Most implementations of malloc/free do not return freed memory
	to the operating system, but merely make it available for future
	malloc() calls within the same program.

7.26:	How does free() know how many bytes to free?

A:	The malloc/free implementation remembers the size of each block
	as it is allocated, so it is not necessary to remind it of the
	size when freeing.

7.27:	So can I query the malloc package to find out how big an
	allocated block is?

A:	Unfortunately, there is no standard or portable way.
	(Some compilers provide nonstandard extensions.)

7.30:	Is it legal to pass a null pointer as the first argument to
	realloc()?  Why would you want to?

A:	ANSI C sanctions this usage (and the related realloc(..., 0),
	which frees), although several earlier implementations do not
	support it, so it may not be fully portable.  Passing an
	initially-null pointer to realloc() can make it easier to write
	a self-starting incremental allocation algorithm.

	References: ISO Sec. 7.10.3.4; H&S Sec. 16.3 p. 388.

7.31:	What's the difference between calloc() and malloc()?  Is it safe
	to take advantage of calloc's zero-filling?  Does free() work
	on memory allocated with calloc(), or do you need a cfree()?

A:	calloc(m, n) is essentially equivalent to

		p = malloc(m * n);
		memset(p, 0, m * n);

	The zero fill is all-bits-zero, and does *not* therefore
	guarantee useful null pointer values (see section 5 of this
	list) or floating-point zero values.  free() is properly used to
	free the memory allocated by calloc().

	References: ISO Sec. 7.10.3 to 7.10.3.2; H&S Sec. 16.1 p. 386,
	Sec. 16.2 p. 386; PCS Sec. 11 pp. 141,142.

7.32:	What is alloca() and why is its use discouraged?

A:	alloca() allocates memory which is automatically freed when the
	function which called alloca() returns.  That is, memory
	allocated with alloca is local to a particular function's "stack
	frame" or context.

	alloca() cannot be written portably, and is difficult to
	implement on machines without a conventional stack.  Its use is
	problematical (and the obvious implementation on a stack-based
	machine fails) when its return value is passed directly to
	another function, as in fgets(alloca(100), 100, stdin).

	For these reasons, alloca() is not Standard and cannot be used
	in programs which must be widely portable, no matter how useful
	it might be.  Now that C99 supports variable-length arrays
	(VLA's), they can be used to more cleanly accomplish most of the
	tasks which alloca() used to be put to.

	See also question 7.22.

	References: Rationale Sec. 4.10.3.


Section 8. Characters and Strings

8.1:	Why doesn't

		strcat(string, '!');

	work?

A:	There is a very real difference between characters and strings,
	and strcat() concatenates *strings*.

	Characters in C are represented by small integers corresponding
	to their character set values (see also question 8.6 below).
	Strings are represented by arrays of characters; you usually
	manipulate a pointer to the first character of the array.  It is
	never correct to use one when the other is expected.  To append
	a ! to a string, use

		strcat(string, "!");

	See also questions 1.32, 7.2, and 16.6.

	References: CT&P Sec. 1.5 pp. 9-10.

8.2:	I'm checking a string to see if it matches a particular value.
	Why isn't this code working?

		char *string;
		...
		if(string == "value") {
			/* string matches "value" */
			...
		}

A:	Strings in C are represented as arrays of characters, and C
	never manipulates (assigns, compares, etc.) arrays as a whole.
	The == operator in the code fragment above compares two pointers
	-- the value of the pointer variable string and a pointer to the
	string literal "value" -- to see if they are equal, that is, if
	they point to the same place.  They probably don't, so the
	comparison never succeeds.

	To compare two strings, you generally use the library function
	strcmp():

		if(strcmp(string, "value") == 0) {
			/* string matches "value" */
			...
		}

8.3:	If I can say

		char a[] = "Hello, world!";

	why can't I say

		char a[14];
		a = "Hello, world!";

A:	Strings are arrays, and you can't assign arrays directly.  Use
	strcpy() instead:

		strcpy(a, "Hello, world!");

	See also questions 1.32, 4.2, and 7.2.

8.6:	How can I get the numeric (character set) value corresponding to
	a character, or vice versa?

A:	In C, characters are represented by small integers corresponding
	to their values in the machine's character set.  Therefore, you
	don't need a conversion function: if you have the character, you
	have its value.

	To convert back and forth between the digit characters and the
	corresponding integers in the range 0-9, add or subtract the
	constant '0' (that is, the character value '0').

	See also questions 13.1 and 20.10.

8.9:	I think something's wrong with my compiler: I just noticed that
	sizeof('a') is 2, not 1 (i.e. not sizeof(char)).

A:	Perhaps surprisingly, character constants in C are of type int,
	so sizeof('a') is sizeof(int) (though this is another area
	where C++ differs).  See also question 7.8.

	References: ISO Sec. 6.1.3.4; H&S Sec. 2.7.3 p. 29.


Section 9. Boolean Expressions and Variables

9.1:	What is the right type to use for Boolean values in C?  Why
	isn't it a standard type?  Should I use #defines or enums for
	the true and false values?

A:	C does not provide a standard Boolean type, in part because
	picking one involves a space/time tradeoff which can best be
	decided by the programmer.  (Using an int may be faster, while
	using char may save data space.  Smaller types may make the
	generated code bigger or slower, though, if they require lots of
	conversions to and from int.)

	The choice between #defines and enumeration constants for the
	true/false values is arbitrary and not terribly interesting (see
	also questions 2.22 and 17.10).  Use any of

		#define TRUE  1			#define YES 1
		#define FALSE 0			#define NO  0

		enum bool {false, true};	enum bool {no, yes};

	or use raw 1 and 0, as long as you are consistent within one
	program or project.  (An enumeration may be preferable if your
	debugger shows the names of enumeration constants when examining
	variables.)

	Some people prefer variants like

		#define TRUE (1==1)
		#define FALSE (!TRUE)

	or define "helper" macros such as

		#define Istrue(e) ((e) != 0)

	These don't buy anything (see question 9.2 below; see also
	questions 5.12 and 10.2).

9.2:	Isn't #defining TRUE to be 1 dangerous, since any nonzero value
	is considered "true" in C?  What if a built-in logical or
	relational operator "returns" something other than 1?

A:	It is true (sic) that any nonzero value is considered true in C,
	but this applies only "on input", i.e. where a Boolean value is
	expected.  When a Boolean value is generated by a built-in
	operator, it is guaranteed to be 1 or 0.  Therefore, the test

		if((a == b) == TRUE)

	would work as expected (as long as TRUE is 1), but it is
	obviously silly.  In fact, explicit tests against TRUE and
	FALSE are generally inappropriate, because some library
	functions (notably isupper(), isalpha(), etc.) return,
	on success, a nonzero value which is not necessarily 1.
	(Besides, if you believe that "if((a == b) == TRUE)" is an
	improvement over "if(a == b)", why stop there?  Why not use
	"if(((a == b) == TRUE) == TRUE)"?)  A good rule of thumb is
	to use TRUE and FALSE (or the like) only for assignment to a
	Boolean variable or function parameter, or as the return value
	from a Boolean function, but never in a comparison.


	The preprocessor macros TRUE and FALSE (and, of course, NULL)
	are used for code readability, not because the underlying values
	might ever change.  (See also questions 5.3 and 5.10.)

	Although the use of macros like TRUE and FALSE (or YES
	and NO) seems clearer, Boolean values and definitions can
	be sufficiently confusing in C that some programmers feel that
	TRUE and FALSE macros only compound the confusion, and prefer
	to use raw 1 and 0 instead.  (See also question 5.9.)

	References: K&R1 Sec. 2.6 p. 39, Sec. 2.7 p. 41; K&R2 Sec. 2.6
	p. 42, Sec. 2.7 p. 44, Sec. A7.4.7 p. 204, Sec. A7.9 p. 206; ISO
	Sec. 6.3.3.3, Sec. 6.3.8, Sec. 6.3.9, Sec. 6.3.13, Sec. 6.3.14,
	Sec. 6.3.15, Sec. 6.6.4.1, Sec. 6.6.5; H&S Sec. 7.5.4 pp. 196-7,
	Sec. 7.6.4 pp. 207-8, Sec. 7.6.5 pp. 208-9, Sec. 7.7 pp. 217-8,
	Sec. 7.8 pp. 218-9, Sec. 8.5 pp. 238-9, Sec. 8.6 pp. 241-4;
	"What the Tortoise Said to Achilles".

9.3:	Is if(p), where p is a pointer, a valid conditional?

A:	Yes.  See question 5.3.


Section 10. C Preprocessor

10.2:	Here are some cute preprocessor macros:

		#define begin	{
		#define end	}

	What do y'all think?

A:	Bleah.  See also section 17.

10.3:	How can I write a generic macro to swap two values?

A:	There is no good answer to this question.  If the values are
	integers, a well-known trick using exclusive-OR could perhaps
	be used, but it will not work for floating-point values or
	pointers, or if the two values are the same variable.  (See
	questions 3.3b and 20.15c.)  If the macro is intended to be
	used on values of arbitrary type (the usual goal), it cannot
	use a temporary, since it does not know what type of temporary
	it needs (and would have a hard time picking a name for it if
	it did), and standard C does not provide a typeof operator.

	The best all-around solution is probably to forget about using a
	macro, unless you're willing to pass in the type as a third
	argument.

10.4:	What's the best way to write a multi-statement macro?

A:	The usual goal is to write a macro that can be invoked as if it
	were a statement consisting of a single function call.  This
	means that the "caller" will be supplying the final semicolon,
	so the macro body should not.  The macro body cannot therefore
	be a simple brace-enclosed compound statement, because syntax
	errors would result if it were invoked (apparently as a single
	statement, but with a resultant extra semicolon) as the if
	branch of an if/else statement with an explicit else clause.

	The traditional solution, therefore, is to use

		#define MACRO(arg1, arg2) do {	/
			/* declarations */	/
			stmt1;			/
			stmt2;			/
			/* ... */		/
			} while(0)	/* (no trailing ; ) */

	When the caller appends a semicolon, this expansion becomes a
	single statement regardless of context.  (An optimizing compiler
	will remove any "dead" tests or branches on the constant
	condition 0, although lint may complain.)

	If all of the statements in the intended macro are simple
	expressions, with no declarations or loops, another technique is
	to write a single, parenthesized expression using one or more
	comma operators.  (For an example, see the first DEBUG() macro
	in question 10.26.)  This technique also allows a value to be
	"returned."

	References: H&S Sec. 3.3.2 p. 45; CT&P Sec. 6.3 pp. 82-3.

10.6:	I'm splitting up a program into multiple source files for the
	first time, and I'm wondering what to put in .c files and what
	to put in .h files.  (What does ".h" mean, anyway?)

A:	As a general rule, you should put these things in header (.h)
	files:

		macro definitions (preprocessor #defines)
		structure, union, and enumeration declarations
		typedef declarations
		external function declarations (see also question 1.11)
		global variable declarations

	It's especially important to put a declaration or definition in
	a header file when it will be shared between several other
	files.  (In particular, never put external function prototypes
	in .c files.  See also question 1.7.)

	On the other hand, when a definition or declaration should
	remain private to one .c file, it's fine to leave it there.

	See also questions 1.7 and 10.7.

	References: K&R2 Sec. 4.5 pp. 81-2; H&S Sec. 9.2.3 p. 267; CT&P
	Sec. 4.6 pp. 66-7.

10.7:	Is it acceptable for one header file to #include another?

A:	It's a question of style, and thus receives considerable debate.
	Many people believe that "nested #include files" are to be
	avoided: the prestigious Indian Hill Style Guide (see question
	17.9) disparages them; they can make it harder to find relevant
	definitions; they can lead to multiple-definition errors if a
	file is #included twice; they can lead to increased compilation
	time; and they make manual Makefile maintenance very difficult.
	On the other hand, they make it possible to use header files in
	a modular way (a header file can #include what it needs itself,
	rather than requiring each #includer to do so); a tool like grep
	(or a tags file) makes it easy to find definitions no matter
	where they are; a popular trick along the lines of:

		#ifndef HFILENAME_USED
		#define HFILENAME_USED
		...header file contents...
		#endif

	(where a different bracketing macro name is used for each header
	file) makes a header file "idempotent" so that it can safely be
	#included multiple times; and automated Makefile maintenance
	tools (which are a virtual necessity in large projects anyway;
	see question 18.1) handle dependency generation in the face of
	nested #include files easily.  See also question 17.10.

	References: Rationale Sec. 4.1.2.

10.8a:	What's the difference between #include <> and #include "" ?

A:	The <> syntax is typically used with Standard or system-supplied
	headers, while "" is typically used for a program's own header
	files.

10.8b:	What are the complete rules for header file searching?

A:	The exact behavior is implementation-defined (which means that
	it is supposed to be documented; see question 11.33).
	Typically, headers named with <> syntax are searched for in one
	or more standard places.  Header files named with "" syntax are
	first searched for in the "current directory," then (if not
	found) in the same standard places.

	Traditionally (especially under Unix compilers), the current
	directory is taken to be the directory containing the file
	containing the #include directive.  Under other compilers,
	however, the current directory (if any) is the directory in
	which the compiler was initially invoked.  Check your compiler
	documentation.

	References: K&R2 Sec. A12.4 p. 231; ISO Sec. 6.8.2; H&S Sec. 3.4
	p. 55.

10.9:	I'm getting strange syntax errors on the very first declaration
	in a file, but it looks fine.

A:	Perhaps there's a missing semicolon at the end of the last
	declaration in the last header file you're #including.  See also
	questions 2.18, 11.29, and 16.1b.

10.10b:	I'm #including the right header file for the library function
	I'm using, but the linker keeps saying it's undefined.

A:	See question 13.25.

10.11:	I'm compiling a program, and I seem to be missing one of the
	header files it requires.  Can someone send me a copy?

A:	There are several situations, depending on what sort of header
	file it is that's "missing".

	If the missing header file is a standard one, there's a problem
	with your compiler.  You'll need to contact your vendor, or
	someone knowledgeable about your particular compiler, for help.

	The situation is more complicated in the case of nonstandard
	headers.  Some are completely system- or compiler-specific.
	Some are completely unnecessary, and should be replaced by their
	Standard equivalents.  (For example, instead of , use
	.)  Other headers, such as those associated with
	popular add-on libraries, may be reasonably portable.

	Standard headers exist in part so that definitions appropriate
	to your compiler, operating system, and processor can be
	supplied.  You cannot just pick up a copy of someone else's
	header file and expect it to work, unless that person is using
	exactly the same environment.  You may actually have a
	portability problem (see section 19), or a compiler problem.
	Otherwise, see question 18.16.

10.12:	How can I construct preprocessor #if expressions which compare
	strings?

A:	You can't do it directly; preprocessor #if arithmetic uses only
	integers.  An alternative is to #define several macros with
	symbolic names and distinct integer values, and implement
	conditionals on those.

	See also question 20.17.

	References: K&R2 Sec. 4.11.3 p. 91; ISO Sec. 6.8.1; H&S
	Sec. 7.11.1 p. 225.

10.13:	Does the sizeof operator work in preprocessor #if directives?

A:	No.  Preprocessing happens during an earlier phase of
	compilation, before type names have been parsed.  Instead of
	sizeof, consider using the predefined constants in ANSI's
	, if applicable, or perhaps a "configure" script.
	(Better yet, try to write code which is inherently insensitive
	to type sizes; see also question 1.1.)

	References: ISO Sec. 5.1.1.2, Sec. 6.8.1; H&S Sec. 7.11.1 p.
	225.

10.14:	Can I use an #ifdef in a #define line, to define something two
	different ways?

A:	No.  You can't "run the preprocessor on itself," so to speak.
	What you can do is use one of two completely separate #define
	lines, depending on the #ifdef setting.

	References: ISO Sec. 6.8.3, Sec. 6.8.3.4; H&S Sec. 3.2 pp. 40-1.

10.15:	Is there anything like an #ifdef for typedefs?

A:	Unfortunately, no.  You may have to keep sets of preprocessor
	macros (e.g. MY_TYPE_DEFINED) recording whether certain typedefs
	have been declared.  (See also question 10.13.)

	References: ISO Sec. 5.1.1.2, Sec. 6.8.1; H&S Sec. 7.11.1 p.
	225.

10.16:	How can I use a preprocessor #if expression to tell if a machine
	is big-endian or little-endian?

A:	You probably can't.  (Preprocessor arithmetic uses only long
	integers, and there is no concept of addressing.)  Are you
	sure you need to know the machine's endianness explicitly?
	Usually it's better to write code which doesn't care.
	See also question 20.9.

	References: ISO Sec. 6.8.1; H&S Sec. 7.11.1 p. 225.

10.18:	I inherited some code which contains far too many #ifdef's for
	my taste.  How can I preprocess the code to leave only one
	conditional compilation set, without running it through the
	preprocessor and expanding all of the #include's and #define's
	as well?

A:	There are programs floating around called unifdef, rmifdef,
	and scpp ("selective C preprocessor") which do exactly this.
	See question 18.16.

10.19:	How can I list all of the predefined identifiers?

A:	There's no standard way, although it is a common need.  gcc
	provides a -dM option which works with -E, and other compilers
	may provide something similar.  If the compiler documentation
	is unhelpful, the most expedient way is probably to extract
	printable strings from the compiler or preprocessor executable
	with something like the Unix strings utility.  Beware that many
	traditional system-specific predefined identifiers (e.g. "unix")
	are non-Standard (because they clash with the user's namespace)
	and are being removed or renamed.

10.20:	I have some old code that tries to construct identifiers with a
	macro like

		#define Paste(a, b) a/**/b

	but it doesn't work any more.

A:	It was an undocumented feature of some early preprocessor
	implementations (notably Reiser's) that comments disappeared
	entirely and could therefore be used for token pasting.  ANSI
	affirms (as did K&R1) that comments are replaced with white
	space.  However, since the need for pasting tokens was
	demonstrated and real, ANSI introduced a well-defined token-
	pasting operator, ##, which can be used like this:

		#define Paste(a, b) a##b

	See also question 11.17.

	References: ISO Sec. 6.8.3.3; Rationale Sec. 3.8.3.3; H&S
	Sec. 3.3.9 p. 52.

10.22:	Why is the macro

		#define TRACE(n) printf("TRACE: %d/n", n)

	giving me the warning "macro replacement within a string
	literal"?  It seems to be expanding

		TRACE(count);
	as
		printf("TRACE: %d/count", count);

A:	See question 11.18.

10.23-4: I'm having trouble using macro arguments inside string
	literals, using the `#' operator.

A:	See questions 11.17 and 11.18.

10.25:	I've got this tricky preprocessing I want to do and I can't
	figure out a way to do it.

A:	C's preprocessor is not intended as a general-purpose tool.
	(Note also that it is not guaranteed to be available as a
	separate program.)  Rather than forcing it to do something
	inappropriate, consider writing your own little special-purpose
	preprocessing tool, instead.  You can easily get a utility like
	make(1) to run it for you automatically.

	If you are trying to preprocess something other than C, consider
	using a general-purpose preprocessor.  (One older one available
	on most Unix systems is m4.)

10.26:	How can I write a macro which takes a variable number of
	arguments?

A:	One popular trick is to define and invoke the macro with a
	single, parenthesized "argument" which in the macro expansion
	becomes the entire argument list, parentheses and all, for a
	function such as printf():

		#define DEBUG(args) (printf("DEBUG: "), printf args)

		if(n != 0) DEBUG(("n is %d/n", n));

	The obvious disadvantage is that the caller must always remember
	to use the extra parentheses.

	gcc has an extension which allows a function-like macro to
	accept a variable number of arguments, but it's not standard.
	Other possible solutions are to use different macros (DEBUG1,
	DEBUG2, etc.) depending on the number of arguments, or to play
	tricky games with commas:

		#define DEBUG(args) (printf("DEBUG: "), printf(args))
		#define _ ,

		DEBUG("i = %d" _ i);

	C99 introduces formal support for function-like macros with
	variable-length argument lists.  The notation ... can appear at
	the end of the macro "prototype" (just as it does for varargs
	functions), and the pseudomacro __VA_ARGS__ in the macro
	definition is replaced by the variable arguments during
	invocation.

	Finally, you can always use a bona-fide function, which can
	take a variable number of arguments in a well-defined way.
	See questions 15.4 and 15.5.  (If you needed a macro
	replacement, try using a function plus a non-function-like
	macro, e.g. #define printf myprintf .)

	References: C9X Sec. 6.8.3, Sec. 6.8.3.1.


Section 11. ANSI/ISO Standard C

11.1:	What is the "ANSI C Standard?"

A:	In 1983, the American National Standards Institute (ANSI)
	commissioned a committee, X3J11, to standardize the C language.
	After a long, arduous process, including several widespread
	public reviews, the committee's work was finally ratified as ANS
	X3.159-1989 on December 14, 1989, and published in the spring of
	1990.  For the most part, ANSI C standardized existing practice,
	with a few additions from C++ (most notably function prototypes)
	and support for multinational character sets (including the
	controversial trigraph sequences).  The ANSI C standard also
	formalized the C run-time library support routines.

	A year or so later, the Standard was adopted as an international
	standard, ISO/IEC 9899:1990, and this ISO Standard replaced the
	earlier X3.159 even within the United States (where it was known
	as ANSI/ISO 9899-1990 [1992]).  As an ISO Standard, it is
	subject to ongoing revision through the release of Technical
	Corrigenda and Normative Addenda.

	In 1994, Technical Corrigendum 1 (TC1) amended the Standard
	in about 40 places, most of them minor corrections or
	clarifications, and Normative Addendum 1 (NA1) added about 50
	pages of new material, mostly specifying new library functions
	for internationalization.  In 1995, TC2 added a few more minor
	corrections.

	Most recently, a major revision of the Standard, "C99", has been
	completed and adopted.

	Several versions of the Standard, including C99 and the original
	ANSI Standard, have included a "Rationale," explaining many of
	its decisions, and discussing a number of subtle points,
	including several of those covered here.

11.2:	How can I get a copy of the Standard?

A:	An electronic (PDF) copy is available on-line, for US$18, from
	www.ansi.org.  Paper copies are available in the United States
	from

		American National Standards Institute
		11 W. 42nd St., 13th floor
		New York, NY  10036  USA
		(+1) 212 642 4900

	and

		Global Engineering Documents
		15 Inverness Way E
		Englewood, CO  80112  USA
		(+1) 303 397 2715
		(800) 854 7179  (U.S. & Canada)

	In other countries, contact the appropriate national standards
	body, or ISO in Geneva at:

		ISO Sales
		Case Postale 56
		CH-1211 Geneve 20
		Switzerland

	(or see URL http://www.iso.ch or check the comp.std.internat FAQ
	list, Standards.Faq).

	The mistitled _Annotated ANSI C Standard_, with annotations by
	Herbert Schildt, contains most of the text of ISO 9899; it is
	published by Osborne/McGraw-Hill, ISBN 0-07-881952-0, and sells
	in the U.S. for approximately $40.  It has been suggested that
	the price differential between this work and the official
	standard reflects the value of the annotations: they are plagued
	by numerous errors and omissions, and a few pages of the
	Standard itself are missing.  Many people on the net recommend
	ignoring the annotations entirely.  A review of the annotations
	("annotated annotations") by Clive Feather can be found on the
	web at http://www.lysator.liu.se/c/schildt.html .

	The text of the original ANSI Rationale can be obtained by
	anonymous ftp from ftp.uu.net (see question 18.16) in directory
	doc/standards/ansi/X3.159-1989, and is also available on the web
	at http://www.lysator.liu.se/c/rat/title.html .  That Rationale
	has also been printed by Silicon Press, ISBN 0-929306-07-4.

	Public review drafts of C9X were available from ISO/IEC
	JTC1/SC22/WG14's web site, http://www.dkuug.dk/JTC1/SC22/WG14/ .

	See also question 11.2b below.

11.2b:	Where can I get information about updates to the Standard?

A:	You can find information (including C9X drafts) at
	the web sites http://www.lysator.liu.se/c/index.html,
	http://www.dkuug.dk/JTC1/SC22/WG14/, and http://www.dmk.com/ .

11.3:	My ANSI compiler complains about a mismatch when it sees

		extern int func(float);

		int func(x)
		float x;
		{ ...

A:	You have mixed the new-style prototype declaration
	"extern int func(float);" with the old-style definition
	"int func(x) float x;".  It is usually possible to mix the two
	styles (see question 11.4), but not in this case.

	Old C (and ANSI C, in the absence of prototypes, and in
	variable-length argument lists; see question 15.2) "widens"
	certain arguments when they are passed to functions.  floats
	are promoted to double, and characters and short integers are
	promoted to int.  (For old-style function definitions, the
	values are automatically converted back to the corresponding
	narrower types within the body of the called function, if they
	are declared that way there.)

	This problem can be fixed either by using new-style syntax
	consistently in the definition:

		int func(float x) { ... }

	or by changing the new-style prototype declaration to match the
	old-style definition:

		extern int func(double);

	(In this case, it would be clearest to change the old-style
	definition to use double as well, if possible.)

	It is arguably much safer to avoid "narrow" (char, short int,
	and float) function arguments and return types altogether.

	See also question 1.25.

	References: K&R1 Sec. A7.1 p. 186; K&R2 Sec. A7.3.2 p. 202; ISO
	Sec. 6.3.2.2, Sec. 6.5.4.3; Rationale Sec. 3.3.2.2,
	Sec. 3.5.4.3; H&S Sec. 9.2 pp. 265-7, Sec. 9.4 pp. 272-3.

11.4:	Can you mix old-style and new-style function syntax?

A:	Doing so is legal, but requires a certain amount of care (see
	especially question 11.3).  Modern practice, however, is to
	use the prototyped form in both declarations and definitions.
	(The old-style syntax is marked as obsolescent, so official
	support for it may be removed some day.)

	References: ISO Sec. 6.7.1, Sec. 6.9.5; H&S Sec. 9.2.2 pp.
	265-7, Sec. 9.2.5 pp. 269-70.

11.5:	Why does the declaration

		extern int f(struct x *p);

	give me an obscure warning message about "struct x declared
	inside parameter list"?

A:	In a quirk of C's normal block scoping rules, a structure
	declared (or even mentioned) for the first time within a
	prototype cannot be compatible with other structures declared in
	the same source file (it goes out of scope at the end of the
	prototype).

	To resolve the problem, precede the prototype with the vacuous-
	looking declaration

		struct x;

	which places an (incomplete) declaration of struct x at file
	scope, so that all following declarations involving struct x can
	at least be sure they're referring to the same struct x.

	References: ISO Sec. 6.1.2.1, Sec. 6.1.2.6, Sec. 6.5.2.3.

11.8:	I don't understand why I can't use const values in initializers
	and array dimensions, as in

		const int n = 5;
		int a[n];

A:	The const qualifier really means "read-only"; an object so
	qualified is a run-time object which cannot (normally) be
	assigned to.  The value of a const-qualified object is therefore
	*not* a constant expression in the full sense of the term.  (C
	is unlike C++ in this regard.)  When you need a true compile-
	time constant, use a preprocessor #define (or perhaps an enum).

	References: ISO Sec. 6.4; H&S Secs. 7.11.2,7.11.3 pp. 226-7.

11.8b:	If you can't modify string literals, why aren't they defined as
	being arrays of const characters?

A:	One reason is that so very much code contains lines like

		char *p = "Hello, world!";

	which are not necessarily incorrect.  These lines would suffer
	the diagnostic messages, but it's really any later attempt to
	modify what p points to which would be problems.

	See also question 1.32.

11.9:	What's the difference between "const char *p" and
	"char * const p"?

A:	"const char *p" (which can also be written "char const *p")
	declares a pointer to a constant character (you can't change any
	pointed-to characters); "char * const p" declares a constant
	pointer to a (variable) character (i.e. you can't change the
	pointer).

	Read these "inside out" to understand them; see also question
	1.21.

	References: ISO Sec. 6.5.4.1; Rationale Sec. 3.5.4.1; H&S
	Sec. 4.4.4 p. 81.

11.10:	Why can't I pass a char ** to a function which expects a
	const char **?

A:	You can use a pointer-to-T (for any type T) where a pointer-to-
	const-T is expected.  However, the rule (an explicit exception)
	which permits slight mismatches in qualified pointer types is
	not applied recursively, but only at the top level.

	If you must assign or pass pointers which have qualifier
	mismatches at other than the first level of indirection, you
	must use explicit casts (e.g. (const char **) in this case),
	although as always, the need for such a cast may indicate a
	deeper problem which the cast doesn't really fix.

	References: ISO Sec. 6.1.2.6, Sec. 6.3.16.1, Sec. 6.5.3; H&S
	Sec. 7.9.1 pp. 221-2.

11.12a:	What's the correct declaration of main()?

A:	Either int main(), int main(void), or int main(int argc,
	char *argv[]) (with alternate spellings of argc and *argv[]
	obviously allowed).  See also questions 11.12b to 11.15 below.

	References: ISO Sec. 5.1.2.2.1, Sec. G.5.1; H&S Sec. 20.1 p.
	416; CT&P Sec. 3.10 pp. 50-51.

11.12b:	Can I declare main() as void, to shut off these annoying
	"main returns no value" messages?

A:	No.  main() must be declared as returning an int, and as
	taking either zero or two arguments, of the appropriate types.
	If you're calling exit() but still getting warnings, you may
	have to insert a redundant return statement (or use some kind
	of "not reached" directive, if available).

	Declaring a function as void does not merely shut off or
	rearrange warnings: it may also result in a different function
	call/return sequence, incompatible with what the caller (in
	main's case, the C run-time startup code) expects.

	(Note that this discussion of main() pertains only to "hosted"
	implementations; none of it applies to "freestanding"
	implementations, which may not even have main().  However,
	freestanding implementations are comparatively rare, and if
	you're using one, you probably know it.  If you've never heard
	of the distinction, you're probably using a hosted
	implementation, and the above rules apply.)

	References: ISO Sec. 5.1.2.2.1, Sec. G.5.1; H&S Sec. 20.1 p.
	416; CT&P Sec. 3.10 pp. 50-51.

11.13:	But what about main's third argument, envp?

A:	It's a non-standard (though common) extension.  If you really
	need to access the environment in ways beyond what the standard
	getenv() function provides, though, the global variable environ
	is probably a better avenue (though it's equally non-standard).

	References: ISO Sec. G.5.1; H&S Sec. 20.1 pp. 416-7.

11.14a:	I believe that declaring void main() can't fail, since I'm
	calling exit() instead of returning, and anyway my operating
	system ignores a program's exit/return status.

A:	It doesn't matter whether main() returns or not, or whether
	anyone looks at the status; the problem is that when main() is
	misdeclared, its caller (the runtime startup code) may not even
	be able to *call* it correctly (due to the potential clash of
	calling conventions; see question 11.12b).

	Your operating system may ignore the exit status, and
	void main() may work for you, but it is not portable and not
	correct.

11.14b:	So what could go wrong?  Are there really any systems where
	void main() doesn't work?

A:	It has been reported that programs using void main() and
	compiled using BC++ 4.5 can crash.  Some compilers (including
	DEC C V4.1 and gcc with certain warnings enabled) will complain
	about void main().

11.15:	The book I've been using, _C Programing for the Compleat Idiot_,
	always uses void main().

A:	Perhaps its author counts himself among the target audience.
	Many books unaccountably use void main() in examples, and assert
	that it's correct.  They're wrong.

11.16:	Is exit(status) truly equivalent to returning the same status
	from main()?

A:	Yes and no.  The Standard says that they are equivalent.
	However, a return from main() cannot be expected to work if
	data local to main() might be needed during cleanup; see also
	question 16.4.  A few very old, nonconforming systems may once
	have had problems with one or the other form.  (Finally, the
	two forms are obviously not equivalent in a recursive call to
	main().)

	References: K&R2 Sec. 7.6 pp. 163-4; ISO Sec. 5.1.2.2.3.

11.17:	I'm trying to use the ANSI "stringizing" preprocessing operator
	`#' to insert the value of a symbolic constant into a message,
	but it keeps stringizing the macro's name rather than its value.

A:	You can use something like the following two-step procedure to
	force a macro to be expanded as well as stringized:

		#define Str(x) #x
		#define Xstr(x) Str(x)
		#define OP plus
		char *opname = Xstr(OP);

	This code sets opname to "plus" rather than "OP".

	An equivalent circumlocution is necessary with the token-pasting
	operator ## when the values (rather than the names) of two
	macros are to be concatenated.

	References: ISO Sec. 6.8.3.2, Sec. 6.8.3.5.

11.18:	What does the message "warning: macro replacement within a
	string literal" mean?

A:	Some pre-ANSI compilers/preprocessors interpreted macro
	definitions like

		#define TRACE(var, fmt) printf("TRACE: var = fmt/n", var)

	such that invocations like

		TRACE(i, %d);

	were expanded as

		printf("TRACE: i = %d/n", i);

	In other words, macro parameters were expanded even inside
	string literals and character constants.

	Macro expansion is *not* defined in this way by K&R or by
	Standard C.  When you do want to turn macro arguments into
	strings, you can use the new # preprocessing operator, along
	with string literal concatenation (another new ANSI feature):

		#define TRACE(var, fmt) /
			printf("TRACE: " #var " = " #fmt "/n", var)

	See also question 11.17 above.

	References: H&S Sec. 3.3.8 p. 51.

11.19:	I'm getting strange syntax errors inside lines I've #ifdeffed
	out.

A:	Under ANSI C, the text inside a "turned off" #if, #ifdef, or
	#ifndef must still consist of "valid preprocessing tokens."
	This means that the characters " and ' must each be paired just
	as in real C code, and the pairs mustn't cross line boundaries.
	(Note particularly that an apostrophe within a contracted word
	looks like the beginning of a character constant.)  Therefore,
	natural-language comments and pseudocode should always be
	written between the "official" comment delimiters /* and */.
	(But see question 20.20, and also 10.25.)

	References: ISO Sec. 5.1.1.2, Sec. 6.1; H&S Sec. 3.2 p. 40.

11.20:	What are #pragmas and what are they good for?

A:	The #pragma directive provides a single, well-defined "escape
	hatch" which can be used for all sorts of (nonportable)
	implementation-specific controls and extensions: source listing
	control, structure packing, warning suppression (like lint's old
	/* NOTREACHED */ comments), etc.

	References: ISO Sec. 6.8.6; H&S Sec. 3.7 p. 61.

11.21:	What does "#pragma once" mean?  I found it in some header files.

A:	It is an extension implemented by some preprocessors to help
	make header files idempotent; it is equivalent to the #ifndef
	trick mentioned in question 10.7, though less portable.

11.22:	Is char a[3] = "abc"; legal?  What does it mean?

A:	It is legal in ANSI C (and perhaps in a few pre-ANSI systems),
	though useful only in rare circumstances.  It declares an array
	of size three, initialized with the three characters 'a', 'b',
	and 'c', *without* the usual terminating '/0' character.  The
	array is therefore not a true C string and cannot be used with
	strcpy, printf %s, etc.

	Most of the time, you should let the compiler count the
	initializers when initializing arrays (in the case of the
	initializer "abc", of course, the computed size will be 4).

	References: ISO Sec. 6.5.7; H&S Sec. 4.6.4 p. 98.

11.24:	Why can't I perform arithmetic on a void * pointer?

A:	The compiler doesn't know the size of the pointed-to objects.
	Before performing arithmetic, convert the pointer either to
	char * or to the pointer type you're trying to manipulate (but
	see also question 4.5).

	References: ISO Sec. 6.1.2.5, Sec. 6.3.6; H&S Sec. 7.6.2 p. 204.

11.25:	What's the difference between memcpy() and memmove()?

A:	memmove() offers guaranteed behavior if the source and
	destination arguments overlap.  memcpy() makes no such
	guarantee, and may therefore be more efficiently implementable.
	When in doubt, it's safer to use memmove().

	References: K&R2 Sec. B3 p. 250; ISO Sec. 7.11.2.1,
	Sec. 7.11.2.2; Rationale Sec. 4.11.2; H&S Sec. 14.3 pp. 341-2;
	PCS Sec. 11 pp. 165-6.

11.26:	What should malloc(0) do?  Return a null pointer or a pointer to
	0 bytes?

A:	The ANSI/ISO Standard says that it may do either; the behavior
	is implementation-defined (see question 11.33).

	References: ISO Sec. 7.10.3; PCS Sec. 16.1 p. 386.

11.27:	Why does the ANSI Standard place limits on the length and case-
	significance of external identifiers?

A:	The problem is linkers which are under control of neither
	the ANSI/ISO Standard nor the C compiler developers on the
	systems which have them.  The limitation is only that
	identifiers be *significant* in some initial sequence of
	characters, not that they be restricted to that many characters
	in total length.  (The limitation was to six characters in the
	original ANSI Standard, but has been relaxed to 31 in C99.)

	References: ISO Sec. 6.1.2, Sec. 6.9.1; Rationale Sec. 3.1.2;
	C9X Sec. 6.1.2; H&S Sec. 2.5 pp. 22-3.

11.29:	My compiler is rejecting the simplest possible test programs,
	with all kinds of syntax errors.

A:	Perhaps it is a pre-ANSI compiler, unable to accept function
	prototypes and the like.

	See also questions 1.31, 10.9, 11.30, and 16.1b.

11.30:	Why are some ANSI/ISO Standard library functions showing up as
	undefined, even though I've got an ANSI compiler?

A:	It's possible to have a compiler available which accepts ANSI
	syntax, but not to have ANSI-compatible header files or run-time
	libraries installed.  (In fact, this situation is rather common
	when using a non-vendor-supplied compiler such as gcc.)  See
	also questions 11.29, 13.25, and 13.26.

11.31:	Does anyone have a tool for converting old-style C programs to
	ANSI C, or vice versa, or for automatically generating
	prototypes?

A:	Two programs, protoize and unprotoize, convert back and forth
	between prototyped and "old style" function definitions and
	declarations.  (These programs do *not* handle full-blown
	translation between "Classic" C and ANSI C.)  These programs are
	part of the FSF's GNU C compiler distribution; see question
	18.3.

	The unproto program (/pub/unix/unproto5.shar.Z on
	ftp.win.tue.nl) is a filter which sits between the preprocessor
	and the next compiler pass, converting most of ANSI C to
	traditional C on-the-fly.

	The GNU GhostScript package comes with a little program called
	ansi2knr.

	Before converting ANSI C back to old-style, beware that such a
	conversion cannot always be made both safely and automatically.
	ANSI C introduces new features and complexities not found in K&R
	C.  You'll especially need to be careful of prototyped function
	calls; you'll probably need to insert explicit casts.  See also
	questions 11.3 and 11.29.

	Several prototype generators exist, many as modifications to
	lint.  A program called CPROTO was posted to comp.sources.misc
	in March, 1992.  There is another program called "cextract."
	Many vendors supply simple utilities like these with their
	compilers.  See also question 18.16.  (But be careful when
	generating prototypes for old functions with "narrow"
	parameters; see question 11.3.)

11.32:	Why won't the Frobozz Magic C Compiler, which claims to be ANSI
	compliant, accept this code?  I know that the code is ANSI,
	because gcc accepts it.

A:	Many compilers support a few non-Standard extensions, gcc more
	so than most.  Are you sure that the code being rejected doesn't
	rely on such an extension?  It is usually a bad idea to perform
	experiments with a particular compiler to determine properties
	of a language; the applicable standard may permit variations, or
	the compiler may be wrong.  See also question 11.35.

11.33:	People seem to make a point of distinguishing between
	implementation-defined, unspecified, and undefined behavior.
	What's the difference?

A:	Briefly: implementation-defined means that an implementation
	must choose some behavior and document it.  Unspecified means
	that an implementation should choose some behavior, but need not
	document it.  Undefined means that absolutely anything might
	happen.  In no case does the Standard impose requirements; in
	the first two cases it occasionally suggests (and may require a
	choice from among) a small set of likely behaviors.

	Note that since the Standard imposes *no* requirements on the
	behavior of a compiler faced with an instance of undefined
	behavior, the compiler can do absolutely anything.  In
	particular, there is no guarantee that the rest of the program
	will perform normally.  It's perilous to think that you can
	tolerate undefined behavior in a program; see question 3.2 for a
	relatively simple example.

	If you're interested in writing portable code, you can ignore
	the distinctions, as you'll usually want to avoid code that
	depends on any of the three behaviors.

	See also questions 3.9, and 11.34.

	(A fourth defined class of not-quite-precisely-defined behavior,
	without the same stigma attached to it, is "locale-specific".)

	References: ISO Sec. 3.10, Sec. 3.16, Sec. 3.17; Rationale
	Sec. 1.6.

11.33b:	What does it really mean for a program to be "legal" or "valid"
	or "conforming"?

A:	Simply stated, the Standard talks about three kinds of
	conformance: conforming programs, strictly conforming programs,
	and conforming implementations.

	A "conforming program" is one that is accepted by a conforming
	implementation.

	A "strictly conforming program" is one that uses the language
	exactly as specified in the Standard, and that does not depend
	on any implementation-defined, unspecified, or undefined
	behavior.

	A "conforming implementation" is one that does everything the
	Standard says it's supposed to.

	References: ISO Sec. ; Rationale Sec. 1.7.

11.34:	I'm appalled that the ANSI Standard leaves so many issues
	undefined.  Isn't a Standard's whole job to standardize these
	things?

A:	It has always been a characteristic of C that certain constructs
	behaved in whatever way a particular compiler or a particular
	piece of hardware chose to implement them.  This deliberate
	imprecision often allows compilers to generate more efficient
	code for common cases, without having to burden all programs
	with extra code to assure well-defined behavior of cases deemed
	to be less reasonable.  Therefore, the Standard is simply
	codifying existing practice.

	A programming language standard can be thought of as a treaty
	between the language user and the compiler implementor.  Parts
	of that treaty consist of features which the compiler
	implementor agrees to provide, and which the user may assume
	will be available.  Other parts, however, consist of rules which
	the user agrees to follow and which the implementor may assume
	will be followed.  As long as both sides uphold their
	guarantees, programs have a fighting chance of working
	correctly.  If *either* side reneges on any of its commitments,
	nothing is guaranteed to work.

	See also question 11.35.

	References: Rationale Sec. 1.1.

11.35:	People keep saying that the behavior of i = i++ is undefined,
	but I just tried it on an ANSI-conforming compiler, and got the
	results I expected.

A:	A compiler may do anything it likes when faced with undefined
	behavior (and, within limits, with implementation-defined and
	unspecified behavior), including doing what you expect.  It's
	unwise to depend on it, though.  See also questions 7.3b, 11.32,
	11.33, and 11.34.


Section 12. Stdio

12.1:	What's wrong with this code?

		char c;
		while((c = getchar()) != EOF) ...

A:	For one thing, the variable to hold getchar's return value must
	be an int.  getchar() can return all possible character values,
	as well as EOF.  By squeezing getchar's return value into a
	char, either a normal character might be misinterpreted as EOF,
	or the EOF might be altered (particularly if type char is
	unsigned) and so never seen.

	References: K&R1 Sec. 1.5 p. 14; K&R2 Sec. 1.5.1 p. 16; ISO
	Sec. 6.1.2.5, Sec. 7.9.1, Sec. 7.9.7.5; H&S Sec. 5.1.3 p. 116,
	Sec. 15.1, Sec. 15.6; CT&P Sec. 5.1 p. 70; PCS Sec. 11 p. 157.

12.1b:	I have a simple little program that reads characters until EOF,
	but how do I actually *enter* that "EOF" value from the
	keyboard?

A:	It turns out that the value of EOF as seen within your C program
	has essentially nothing to do with the keystroke combination you
	might use to signal end-of-file from the keyboard.  Depending on
	your operating system, you indicate end-of-file from the
	keyboard using various keystroke combinations, usually either
	control-D or control-Z.

12.2:	Why does the code

		while(!feof(infp)) {
			fgets(buf, MAXLINE, infp);
			fputs(buf, outfp);
		}

	copy the last line twice?

A:	In C, end-of-file is only indicated *after* an input routine has
	tried to read, and failed.  (In other words, C's I/O is not like
	Pascal's.)  Usually, you should just check the return value of
	the input routine -- fgets(), for example, returns NULL on end-
	of-file.  In virtually all cases, there's no need to use feof()
	at all.

	References: K&R2 Sec. 7.6 p. 164; ISO Sec. 7.9.3, Sec. 7.9.7.1,
	Sec. 7.9.10.2; H&S Sec. 15.14 p. 382.

12.4:	My program's prompts and intermediate output don't always show
	up on the screen, especially when I pipe the output through
	another program.

A:	It's best to use an explicit fflush(stdout) whenever output
	should definitely be visible (and especially if the text does
	not end with /n).  Several mechanisms attempt to perform the
	fflush() for you, at the "right time," but they tend to apply
	only when stdout is an interactive terminal.  (See also question
	12.24.)

	References: ISO Sec. 7.9.5.2.

12.5:	How can I read one character at a time, without waiting for the
	RETURN key?

A:	See question 19.1.

12.6:	How can I print a '%' character in a printf format string?  I
	tried /%, but it didn't work.

A:	Simply double the percent sign: %% .

	/% can't work, because the backslash / is the *compiler's*
	escape character, while here our problem is that the % is
	essentially printf's escape character.

	See also question 19.17.

	References: K&R1 Sec. 7.3 p. 147; K&R2 Sec. 7.2 p. 154; ISO
	Sec. 7.9.6.1.

12.9:	Someone told me it was wrong to use %lf with printf().  How can
	printf() use %f for type double, if scanf() requires %lf?

A:	It's true that printf's %f specifier works with both float and
	double arguments.  Due to the "default argument promotions"
	(which apply in variable-length argument lists such as printf's,
	whether or not prototypes are in scope), values of type float
	are promoted to double, and printf() therefore sees only
	doubles.  (printf() does accept %Lf, for long double.)
	See also questions 12.13 and 15.2.

	References: K&R1 Sec. 7.3 pp. 145-47, Sec. 7.4 pp. 147-50; K&R2
	Sec. 7.2 pp. 153-44, Sec. 7.4 pp. 157-59; ISO Sec. 7.9.6.1,
	Sec. 7.9.6.2; H&S Sec. 15.8 pp. 357-64, Sec. 15.11 pp. 366-78;
	CT&P Sec. A.1 pp. 121-33.

12.9b:	What printf format should I use for a typedef like size_t
	when I don't know whether it's long or some other type?

A:	Use a cast to convert the value to a known, conservatively-
	sized type, then use the printf format matching that type.
	For example, to print the size of a type, you might use

		printf("%lu", (unsigned long)sizeof(thetype));

12.10:	How can I implement a variable field width with printf?
	That is, instead of %8d, I want the width to be specified
	at run time.

A:	printf("%*d", width, x) will do just what you want.
	See also question 12.15.

	References: K&R1 Sec. 7.3; K&R2 Sec. 7.2; ISO Sec. 7.9.6.1; H&S
	Sec. 15.11.6; CT&P Sec. A.1.

12.11:	How can I print numbers with commas separating the thousands?
	What about currency formatted numbers?

A:	The functions in  begin to provide some support for
	these operations, but there is no standard function for doing
	either task.  (The only thing printf() does in response to a
	custom locale setting is to change its decimal-point character.)

	References: ISO Sec. 7.4; H&S Sec. 11.6 pp. 301-4.

12.12:	Why doesn't the call scanf("%d", i) work?

A:	The arguments you pass to scanf() must always be pointers.
	To fix the fragment above, change it to scanf("%d", &i) .

12.12b:	Why *does* the call

		char s[30];
		scanf("%s", s);

	work (without the &)?

A:	You always need a *pointer*; you don't necessarily need an
	explicit &.  When you pass an array to scanf(), you do not need
	the &, because arrays are always passed to functions as
	pointers, whether you use & or not.  See questions 6.3 and 6.4.

12.13:	Why doesn't this code:

		double d;
		scanf("%f", &d);

	work?

A:	Unlike printf(), scanf() uses %lf for values of type double, and
	%f for float.  See also question 12.9.

12.15:	How can I specify a variable width in a scanf() format string?

A:	You can't; an asterisk in a scanf() format string means to
	suppress assignment.  You may be able to use ANSI stringizing
	and string concatenation to accomplish about the same thing, or
	you can construct the scanf format string at run time.

12.17:	When I read numbers from the keyboard with scanf "%d/n", it
	seems to hang until I type one extra line of input.

A:	Perhaps surprisingly, /n in a scanf format string does *not*
	mean to expect a newline, but rather to read and discard
	characters as long as each is a whitespace character.
	See also question 12.20.

	References: K&R2 Sec. B1.3 pp. 245-6; ISO Sec. 7.9.6.2; H&S
	Sec. 15.8 pp. 357-64.

12.18a:	I'm reading a number with scanf %d and then a string with
	gets(), but the compiler seems to be skipping the call to
	gets()!

A:	scanf %d won't consume a trailing newline.  If the input number
	is immediately followed by a newline, that newline will
	immediately satisfy the gets().

	As a general rule, you shouldn't try to interlace calls to
	scanf() with calls to gets() (or any other input routines);
	scanf's peculiar treatment of newlines almost always leads to
	trouble.  Either use scanf() to read everything or nothing.

	See also questions 12.20 and 12.23.

	References: ISO Sec. 7.9.6.2; H&S Sec. 15.8 pp. 357-64.

12.19:	I figured I could use scanf() more safely if I checked its
	return value to make sure that the user typed the numeric values
	I expect, but sometimes it seems to go into an infinite loop.

A:	When scanf() is attempting to convert numbers, any non-numeric
	characters it encounters terminate the conversion *and are left
	on the input stream*.  Therefore, unless some other steps are
	taken, unexpected non-numeric input "jams" scanf() again and
	again: scanf() never gets past the bad character(s) to encounter
	later, valid data.  If the user types a character like `x' in
	response to a numeric scanf format such as %d or %f, code that
	simply re-prompts and retries the same scanf() call will
	immediately reencounter the same `x'.

	See also question 12.20.

	References: ISO Sec. 7.9.6.2; H&S Sec. 15.8 pp. 357-64.

12.20:	Why does everyone say not to use scanf()?  What should I use
	instead?

A:	scanf() has a number of problems -- see questions 12.17, 12.18a,
	and 12.19.  Also, its %s format has the same problem that gets()
	has (see question 12.23) -- it's hard to guarantee that the
	receiving buffer won't overflow.

	More generally, scanf() is designed for relatively structured,
	formatted input (its name is in fact derived from "scan
	formatted").  If you pay attention, it will tell you whether it
	succeeded or failed, but it can tell you only approximately
	where it failed, and not at all how or why.  It's nearly
	impossible to do decent error recovery with scanf(); usually
	it's far easier to read entire lines (with fgets() or the like),
	then interpret them, either using sscanf() or some other
	techniques.  (Functions like strtol(), strtok(), and atoi() are
	often useful; see also question 13.6.)  If you do use any scanf
	variant, be sure to check the return value to make sure that the
	expected number of items were found.  Also, if you use %s, be
	sure to guard against buffer overflow.

	References: K&R2 Sec. 7.4 p. 159.

12.21:	How can I tell how much destination buffer space I'll need for
	an arbitrary sprintf call?  How can I avoid overflowing the
	destination buffer with sprintf()?

A:	When the format string being used with sprintf() is known and
	relatively simple, you can sometimes predict a buffer size in an
	ad-hoc way.  If the format consists of one or two %s's, you can
	count the fixed characters in the format string yourself (or let
	sizeof count them for you) and add in the result of calling
	strlen() on the string(s) to be inserted.  For integers, the
	number of characters produced by %d is no more than

		((sizeof(int) * CHAR_BIT + 2) / 3 + 1)	/* +1 for '-' */

	(CHAR_BIT is in ), though this computation may be
	over-conservative.  (It computes the number of characters
	required for a base-8 representation of a number; a base-10
	expansion is guaranteed to take as much room or less.)

	When the format string is more complicated, or is not even known
	until run time, predicting the buffer size becomes as difficult
	as reimplementing sprintf(), and correspondingly error-prone
	(and inadvisable).  A last-ditch technique which is sometimes
	suggested is to use fprintf() to print the same text to a bit
	bucket or temporary file, and then to look at fprintf's return
	value or the size of the file (but see question 19.12, and worry
	about write errors).

	If there's any chance that the buffer might not be big enough,
	you won't want to call sprintf() without some guarantee that the
	buffer will not overflow and overwrite some other part of
	memory.  If the format string is known, you can limit %s
	expansion by using %.Ns for some N, or %.*s (see also question
	12.10).

	To avoid the overflow problem, you can use a length-limited
	version of sprintf(), namely snprintf().  It is used like this:

		snprintf(buf, bufsize, "You typed /"%s/"", answer);

	snprintf() has been available in several stdio libraries
	(including GNU and 4.4bsd) for several years.  It has finally
	been standardized in C99.

	As an extra, added bonus, the C99 snprintf() provides a way
	to predict the size required for an arbitrary sprintf() call.
	C99's snprintf() returns the number of characters it would have
	placed in the buffer, and it may be called with a buffer size
	of 0.  Therefore, the call

		nch = snprintf(NULL, 0, fmtstring, /* other arguments */ );

	predicts the number of characters required for the fully-
	formatted string.

	Yet another option is the (nonstandard) asprintf() function,
	present in various C libraries including bsd's and GNU's, which
	formats to (and returns a pointer to) a malloc'ed buffer, like
	this:

		char *buf;
		asprintf(&buf, "%d = %s", 42, "forty-two");
		/* now buf points to malloc'ed space containing formatted string */

	References: C9X Sec. 7.13.6.6.

12.23:	Why does everyone say not to use gets()?

A:	Unlike fgets(), gets() cannot be told the size of the buffer
	it's to read into, so it cannot be prevented from overflowing
	that buffer.  The Standard fgets() function is a vast
	improvement over gets(), although it's not perfect, either.
	(If long lines are a real possibility, their proper handling
	must be carefully considered.)  See question 7.1 for a code
	fragment illustrating the replacement of gets() with fgets().

	References: Rationale Sec. 4.9.7.2; H&S Sec. 15.7 p. 356.

12.24:	Why does errno contain ENOTTY after a call to printf()?

A:	Many implementations of the stdio package adjust their behavior
	slightly if stdout is a terminal.  To make the determination,
	these implementations perform some operation which happens to
	fail (with ENOTTY) if stdout is not a terminal.  Although the
	output operation goes on to complete successfully, errno still
	contains ENOTTY.  (Note that it is only meaningful for a program
	to inspect the contents of errno after an error has been
	reported; errno is not guaranteed to be 0 otherwise.)

	References: ISO Sec. 7.1.4, Sec. 7.9.10.3; CT&P Sec. 5.4 p. 73;
	PCS Sec. 14 p. 254.

12.25:	What's the difference between fgetpos/fsetpos and ftell/fseek?
	What are fgetpos() and fsetpos() good for?

A:	ftell() and fseek() use type long int to represent offsets
	(positions) in a file, and may therefore be limited to offsets
	of about 2 billion (2**31-1).  The newer fgetpos() and fsetpos()
	functions, on the other hand, use a special typedef, fpos_t, to
	represent the offsets.  The type behind this typedef, if chosen
	appropriately, can represent arbitrarily large offsets, so
	fgetpos() and fsetpos() can be used with arbitrarily huge files.
	fgetpos() and fsetpos() also record the state associated with
	multibyte streams.  See also question 1.4.

	References: K&R2 Sec. B1.6 p. 248; ISO Sec. 7.9.1,
	Secs. 7.9.9.1,7.9.9.3; H&S Sec. 15.5 p. 252.

12.26a:	How can I flush pending input so that a user's typeahead isn't
	read at the next prompt?  Will fflush(stdin) work?

A:	fflush() is defined only for output streams.  Since its
	definition of "flush" is to complete the writing of buffered
	characters (not to discard them), discarding unread input would
	not be an analogous meaning for fflush on input streams.
	See also question 12.26b.

	References: ISO Sec. 7.9.5.2; H&S Sec. 15.2.

12.26b:	If fflush() won't work, what can I use to flush input?

A:	It depends on what you're trying to do.  If you're trying to get
	rid of an unread newline or other unexpected input after calling
	scanf() (see questions 12.18a-12.19), you really need to rewrite
	or replace the call to scanf() (see question 12.20).
	Alternatively, you can consume the rest of a partially-read line
	with a simple code fragment like

		while((c = getchar()) != '/n' && c != EOF)
			/* discard */ ;

	(You may also be able to use the curses flushinp() function.)

	There is no standard way to discard unread characters from a
	stdio input stream, nor would such a way necessarily be
	sufficient, since unread characters can also accumulate in
	other, OS-level input buffers.  If you're trying to actively
	discard typed-ahead input (perhaps in anticipation of issuing a
	critical prompt), you'll have to use a system-specific
	technique; see questions 19.1 and 19.2.

	References: ISO Sec. 7.9.5.2; H&S Sec. 15.2.

12.27:	fopen() is failing for certain pathnames.

A:	See questions 19.17 and 19.17b.

12.30:	I'm trying to update a file in place, by using fopen mode "r+",
	reading a certain string, and writing back a modified string,
	but it's not working.

A:	Be sure to call fseek before you write, both to seek back to the
	beginning of the string you're trying to overwrite, and because
	an fseek or fflush is always required between reading and
	writing in the read/write "+" modes.  Also, remember that you
	can only overwrite characters with the same number of
	replacement characters, and that overwriting in text mode may
	truncate the file at that point, and that you may have to
	preserve line lengths.  See also question 19.14.

	References: ISO Sec. 7.9.5.3.

12.33:	How can I redirect stdin or stdout to a file from within a
	program?

A:	Use freopen() (but see question 12.34 below).

	References: ISO Sec. 7.9.5.4; H&S Sec. 15.2.

12.34:	Once I've used freopen(), how can I get the original stdout (or
	stdin) back?

A:	There isn't a good way.  If you need to switch back, the best
	solution is not to have used freopen() in the first place.  Try
	using your own explicit output (or input) stream variable, which
	you can reassign at will, while leaving the original stdout (or
	stdin) undisturbed.

	It may be possible, in a nonportable way, to save away
	information about a stream before calling freopen(), such that
	the original stream can later be restored.  One way is to use a
	system-specific call such as dup() or dup2(), if available.
	Another is to copy or inspect the contents of the FILE
	structure, but this is exceedingly nonportable and unreliable.

12.36b:	How can I arrange to have output go two places at once,
	e.g. to the screen and to a file?

A:	You can't do this directly, but you could write your own printf
	variant which printed everything twice.  Here is a simple
	example:

		#include 
		#include 

		void f2printf(FILE *fp1, FILE *fp2, char *fmt, ...)
		{
		    va_list argp;
		    va_start(argp, fmt); vfprintf(fp1, fmt, argp); va_end(argp);
		    va_start(argp, fmt); vfprintf(fp2, fmt, argp); va_end(argp);
		}

	where f2printf() is just like fprintf() except that you give it
	two file pointers and it prints to both of them.

	See also question 15.5.

12.38:	How can I read a binary data file properly?  I'm occasionally
	seeing 0x0a and 0x0d values getting garbled, and I seem to hit
	EOF prematurely if the data contains the value 0x1a.

A:	When you're reading a binary data file, you should specify "rb"
	mode when calling fopen(), to make sure that text file
	translations do not occur.  Similarly, when writing binary data
	files, use "wb".

	Note that the text/binary distinction is made when you open the
	file: once a file is open, it doesn't matter which I/O calls you
	use on it.  See also question 20.5.

	References: ISO Sec. 7.9.5.3; H&S Sec. 15.2.1 p. 348.


Section 13. Library Functions

13.1:	How can I convert numbers to strings (the opposite of atoi)?
	Is there an itoa() function?

A:	Just use sprintf().  (Don't worry that sprintf() may be
	overkill, potentially wasting run time or code space; it works
	well in practice.)  See the examples in the answer to question
	7.5a; see also questions 8.6 and 12.21.

	You can obviously use sprintf() to convert long or floating-
	point numbers to strings as well (using %ld or %f).

	References: K&R1 Sec. 3.6 p. 60; K&R2 Sec. 3.6 p. 64.

13.2:	Why does strncpy() not always place a '/0' terminator in the
	destination string?

A:	strncpy() was first designed to handle a now-obsolete data
	structure, the fixed-length, not-necessarily-/0-terminated
	"string."  (A related quirk of strncpy's is that it pads short
	strings with multiple /0's, out to the specified length.)
	strncpy() is admittedly a bit cumbersome to use in other
	contexts, since you must often append a '/0' to the destination
	string by hand.  You can get around the problem by using
	strncat() instead of strncpy(): if the destination string starts
	out empty (that is, if you do *dest = '/0' first), strncat()
	does what you probably wanted strncpy() to do.  Another
	possibility is sprintf(dest, "%.*s", n, source) .

	When arbitrary bytes (as opposed to strings) are being copied,
	memcpy() is usually a more appropriate function to use than
	strncpy().

13.5:	Why do some versions of toupper() act strangely if given an
	upper-case letter?
	Why does some code call islower() before toupper()?

A:	Older versions of toupper() and tolower() did not always work
	correctly on arguments which did not need converting (i.e. on
	digits or punctuation or letters already of the desired case).
	In ANSI/ISO Standard C, these functions are guaranteed to work
	appropriately on all character arguments.

	References: ISO Sec. 7.3.2; H&S Sec. 12.9 pp. 320-1; PCS p. 182.

13.6:	How can I split up a string into whitespace-separated fields?
	How can I duplicate the process by which main() is handed argc
	and argv?

A:	The only Standard function available for this kind of
	"tokenizing" is strtok(), although it can be tricky to use and
	it may not do everything you want it to.  (For instance, it does
	not handle quoting.)

	References: K&R2 Sec. B3 p. 250; ISO Sec. 7.11.5.8; H&S
	Sec. 13.7 pp. 333-4; PCS p. 178.

13.7:	I need some code to do regular expression and wildcard matching.

A:	Make sure you recognize the difference between classic regular
	expressions (variants of which are used in such Unix utilities
	as ed and grep), and filename wildcards (variants of which are
	used by most operating systems).

	There are a number of packages available for matching regular
	expressions.  Most packages use a pair of functions, one for
	"compiling" the regular expression, and one for "executing" it
	(i.e. matching strings against it).  Look for header files named
	 or , and functions called regcmp/regex,
	regcomp/regexec, or re_comp/re_exec.  (These functions may
	exist in a separate regexp library.)  A popular, freely-
	redistributable regexp package by Henry Spencer is available
	from ftp.cs.toronto.edu in pub/regexp.shar.Z or in several other
	archives.  The GNU project has a package called rx.  See also
	question 18.16.

	Filename wildcard matching (sometimes called "globbing") is done
	in a variety of ways on different systems.  On Unix, wildcards
	are automatically expanded by the shell before a process is
	invoked, so programs rarely have to worry about them explicitly.
	Under MS-DOS compilers, there is often a special object file
	which can be linked in to a program to expand wildcards while
	argv is being built.  Several systems (including MS-DOS and VMS)
	provide system services for listing or opening files specified
	by wildcards.  Check your compiler/library documentation.  See
	also questions 19.20 and 20.3.

13.8:	I'm trying to sort an array of strings with qsort(), using
	strcmp() as the comparison function, but it's not working.

A:	By "array of strings" you probably mean "array of pointers to
	char."  The arguments to qsort's comparison function are
	pointers to the objects being sorted, in this case, pointers to
	pointers to char.  strcmp(), however, accepts simple pointers to
	char.  Therefore, strcmp() can't be used directly.  Write an
	intermediate comparison function like this:

		/* compare strings via pointers */
		int pstrcmp(const void *p1, const void *p2)
		{
			return strcmp(*(char * const *)p1, *(char * const *)p2);
		}

	The comparison function's arguments are expressed as "generic
	pointers," const void *.  They are converted back to what they
	"really are" (pointers to pointers to char) and dereferenced,
	yielding char *'s which can be passed to strcmp().

	(Don't be misled by the discussion in K&R2 Sec. 5.11 pp. 119-20,
	which is not discussing the Standard library's qsort).

	References: ISO Sec. 7.10.5.2; H&S Sec. 20.5 p. 419.

13.9:	Now I'm trying to sort an array of structures with qsort().  My
	comparison function takes pointers to structures, but the
	compiler complains that the function is of the wrong type for
	qsort().  How can I cast the function pointer to shut off the
	warning?

A:	The conversions must be in the comparison function, which must
	be declared as accepting "generic pointers" (const void *) as
	discussed in question 13.8 above.  The comparison function might
	look like

		int mystructcmp(const void *p1, const void *p2)
		{
			const struct mystruct *sp1 = p1;
			const struct mystruct *sp2 = p2;
			/* now compare sp1->whatever and sp2-> ... */

	(The conversions from generic pointers to struct mystruct
	pointers happen in the initializations sp1 = p1 and sp2 = p2;
	the compiler performs the conversions implicitly since p1 and p2
	are void pointers.)

	If, on the other hand, you're sorting pointers to structures,
	you'll need indirection, as in question 13.8:
	sp1 = *(struct mystruct * const *)p1 .

	In general, it is a bad idea to insert casts just to "shut the
	compiler up."  Compiler warnings are usually trying to tell you
	something, and unless you really know what you're doing, you
	ignore or muzzle them at your peril.  See also question 4.9.

	References: ISO Sec. 7.10.5.2; H&S Sec. 20.5 p. 419.

13.10:	How can I sort a linked list?

A:	Sometimes it's easier to keep the list in order as you build it
	(or perhaps to use a tree instead).  Algorithms like insertion
	sort and merge sort lend themselves ideally to use with linked
	lists.  If you want to use a standard library function, you can
	allocate a temporary array of pointers, fill it in with pointers
	to all your list nodes, call qsort(), and finally rebuild the
	list pointers based on the sorted array.

	References: Knuth Sec. 5.2.1 pp. 80-102, Sec. 5.2.4 pp. 159-168;
	Sedgewick Sec. 8 pp. 98-100, Sec. 12 pp. 163-175.

13.11:	How can I sort more data than will fit in memory?

A:	You want an "external sort," which you can read about in Knuth,
	Volume 3.  The basic idea is to sort the data in chunks (as much
	as will fit in memory at one time), write each sorted chunk to a
	temporary file, and then merge the files.  Your operating system
	may provide a general-purpose sort utility, and if so, you can
	try invoking it from within your program: see questions 19.27
	and 19.30.

	References: Knuth Sec. 5.4 pp. 247-378; Sedgewick Sec. 13 pp.
	177-187.

13.12:	How can I get the current date or time of day in a C program?

A:	Just use the time(), ctime(), localtime() and/or strftime()
	functions.  Here is a simple example:

		#include 
		#include 

		int main()
		{
			time_t now;
			time(&now);
			printf("It's %s", ctime(&now));
			return 0;
		}

	If you need control over the format, use strftime().
	If you need sub-second resolution, see question 19.37.

	References: K&R2 Sec. B10 pp. 255-7; ISO Sec. 7.12; H&S Sec. 18.

13.13:	I know that the library function localtime() will convert a
	time_t into a broken-down struct tm, and that ctime() will
	convert a time_t to a printable string.  How can I perform the
	inverse operations of converting a struct tm or a string into a
	time_t?

A:	ANSI C specifies a library function, mktime(), which converts a
	struct tm to a time_t.

	Converting a string to a time_t is harder, because of the wide
	variety of date and time formats which might be encountered.
	Some systems provide a strptime() function, which is basically
	the inverse of strftime().  Other popular functions are partime()
	(widely distributed with the RCS package) and getdate() (and a
	few others, from the C news distribution).  See question 18.16.

	References: K&R2 Sec. B10 p. 256; ISO Sec. 7.12.2.3; H&S
	Sec. 18.4 pp. 401-2.

13.14:	How can I add N days to a date?  How can I find the difference
	between two dates?

A:	The ANSI/ISO Standard C mktime() and difftime() functions
	provide some (limited) support for both problems.  mktime()
	accepts non-normalized dates, so it is straightforward to take a
	filled-in struct tm, add or subtract from the tm_mday field, and
	call mktime() to normalize the year, month, and day fields (and
	incidentally convert to a time_t value).  difftime() computes
	the difference, in seconds, between two time_t values; mktime()
	can be used to compute time_t values for two dates to be
	subtracted.

	However, these solutions are guaranteed to work correctly only
	for dates in the range which can be represented as time_t's.
	(For conservatively-sized time_t, that range is often -- but not
	always -- from 1970 to approximately 2037; note however that
	there are time_t representations other than as specified by Unix
	and Posix.)  The tm_mday field is an int, so day offsets of more
	than 32,736 or so may cause overflow.  Note also that at
	daylight saving time changeovers, local days are not 24 hours
	long (so don't assume that division by 86400 will be exact).

	Another approach to both problems, which will work over a much
	wider range of dates, is to use "Julian day numbers".  Code for
	handling Julian day numbers can be found in the Snippets
	collection (see question 18.15c), the Simtel/Oakland archives
	(file JULCAL10.ZIP, see question 18.16), and the "Date
	conversions" article mentioned in the References.

	See also questions 13.13, 20.31, and 20.32.

	References: K&R2 Sec. B10 p. 256; ISO Secs. 7.12.2.2,7.12.2.3;
	H&S Secs. 18.4,18.5 pp. 401-2; David Burki, "Date Conversions".

13.15:	I need a random number generator.

A:	The Standard C library has one: rand().  The implementation on
	your system may not be perfect, but writing a better one isn't
	necessarily easy, either.

	If you do find yourself needing to implement your own random
	number generator, there is plenty of literature out there; see
	the References below or the sci.math.num-analysis FAQ list.
	There are also any number of packages on the net: old standbys
	are r250, RANLIB, and FSULTRA (see question 18.16), and there is
	much recent work by Marsaglia, and Matumoto and Nishimura (the
	"Mersenne Twister"), and some code collected by Don Knuth on his
	web pages.

	References: K&R2 Sec. 2.7 p. 46, Sec. 7.8.7 p. 168; ISO
	Sec. 7.10.2.1; H&S Sec. 17.7 p. 393; PCS Sec. 11 p. 172; Knuth
	Vol. 2 Chap. 3 pp. 1-177; Park and Miller, "Random Number
	Generators: Good Ones are Hard to Find".

13.16:	How can I get random integers in a certain range?

A:	The obvious way,

		rand() % N		/* POOR */

	(which tries to return numbers from 0 to N-1) is poor, because
	the low-order bits of many random number generators are
	distressingly *non*-random.  (See question 13.18.)  A better
	method is something like

		(int)((double)rand() / ((double)RAND_MAX + 1) * N)

	If you'd rather not use floating point, another method is

		rand() / (RAND_MAX / N + 1)

	Both methods obviously require knowing RAND_MAX (which ANSI
	#defines in ), and assume that N is much less than
	RAND_MAX.

	(Note, by the way, that RAND_MAX is a *constant* telling you
	what the fixed range of the C library rand() function is.  You
	cannot set RAND_MAX to some other value, and there is no way of
	requesting that rand() return numbers in some other range.)

	If you're starting with a random number generator which returns
	floating-point values between 0 and 1, all you have to do to get
	integers from 0 to N-1 is multiply the output of that generator
	by N.

	References: K&R2 Sec. 7.8.7 p. 168; PCS Sec. 11 p. 172.

13.17:	Each time I run my program, I get the same sequence of numbers
	back from rand().

A:	You can call srand() to seed the pseudo-random number generator
	with a truly random (or at least variable) initial value, such
	as the time of day.  Here is a simple example:

		#include 
		#include 

		srand((unsigned int)time((time_t *)NULL));

	(Unfortunately, this code isn't perfect -- among other things,
	the time_t returned by time() might be a floating-point type,
	hence not portably convertible to unsigned int without the
	possibility of overflow.  See also question 19.37.)

	Note also that it's rarely useful to call srand() more than once
	during a run of a program; in particular, don't try calling
	srand() before each call to rand(), in an attempt to get "really
	random" numbers.

	References: K&R2 Sec. 7.8.7 p. 168; ISO Sec. 7.10.2.2; H&S
	Sec. 17.7 p. 393.

13.18:	I need a random true/false value, so I'm just taking rand() % 2,
	but it's alternating 0, 1, 0, 1, 0...

A:	Poor pseudorandom number generators (such as the ones
	unfortunately supplied with some systems) are not very random in
	the low-order bits.  Try using the higher-order bits: see
	question 13.16.

	References: Knuth Sec. 3.2.1.1 pp. 12-14.

13.20:	How can I generate random numbers with a normal or Gaussian
	distribution?

A:	Here is one method, recommended by Knuth and due originally to
	Marsaglia:

		#include 
		#include 

		double gaussrand()
		{
			static double V1, V2, S;
			static int phase = 0;
			double X;

			if(phase == 0) {
				do {
					double U1 = (double)rand() / RAND_MAX;
					double U2 = (double)rand() / RAND_MAX;

					V1 = 2 * U1 - 1;
					V2 = 2 * U2 - 1;
					S = V1 * V1 + V2 * V2;
					} while(S >= 1 || S == 0);

				X = V1 * sqrt(-2 * log(S) / S);
			} else
				X = V2 * sqrt(-2 * log(S) / S);

			phase = 1 - phase;

			return X;
		}

	See the extended versions of this list (see question 20.40) for
	other ideas.

	References: Knuth Sec. 3.4.1 p. 117; Marsaglia and Bray,
	"A Convenient Method for Generating Normal Variables";
	Press et al., _Numerical Recipes in C_ Sec. 7.2 pp. 288-290.

13.25:	I keep getting errors due to library functions being undefined,
	but I'm #including all the right header files.

A:	In general, a header file contains only external declarations.
	In some cases (especially if the functions are nonstandard)
	obtaining the actual *definitions* may require explicitly asking
	for the correct libraries to be searched when you link the
	program.  (#including the header doesn't do that.)  See also
	questions 10.11, 11.30, 13.26, 14.3, and 19.40.

13.26:	I'm still getting errors due to library functions being
	undefined, even though I'm explicitly requesting the right
	libraries while linking.

A:	Many linkers make one pass over the list of object files and
	libraries you specify, and extract from libraries only those
	modules which satisfy references which have so far come up as
	undefined.  Therefore, the order in which libraries are listed
	with respect to object files (and each other) is significant;
	usually, you want to search the libraries last.  (For example,
	under Unix, put any -l options towards the end of the command
	line.)  See also question 13.28.

13.28:	What does it mean when the linker says that _end is undefined?

A:	That message is a quirk of the old Unix linkers.  You get an
	error about _end being undefined only when other symbols are
	undefined, too -- fix the others, and the error about _end will
	disappear.  (See also questions 13.25 and 13.26.)

13.29:	My compiler is complaining that printf is undefined!
	How can this be?

A:	Allegedly, there are C compilers for Microsoft Windows which do
	not support printf().  It may be possible to convince such a
	compiler that what you are writing is a "console application"
	meaning that it will open a "console window" in which printf()
	is supported.


Section 14. Floating Point

14.1:	When I set a float variable to, say, 3.1, why is printf printing
	it as 3.0999999?

A:	Most computers use base 2 for floating-point numbers as well as
	for integers.  Although 0.1 is a nice, polite-looking fraction
	in base 10, its base-2 representation is an infinitely-repeating
	fraction (0.0001100110011...), so exact decimal fractions such
	as 3.1 cannot be represented exactly in binary.  Depending on
	how carefully your compiler's binary/decimal conversion routines
	(such as those used by printf) have been written, you may see
	discrepancies when numbers not exactly representable in base 2
	are assigned or read in and then printed (i.e. converted from
	base 10 to base 2 and back again).  See also question 14.6.

14.2:	I'm trying to take some square roots, but I'm getting crazy
	numbers.

A:	Make sure that you have #included , and correctly
	declared other functions returning double.  (Another library
	function to be careful with is atof(), which is declared in
	.)  See also question 14.3 below.

	References: CT&P Sec. 4.5 pp. 65-6.

14.3:	I'm trying to do some simple trig, and I am #including ,
	but I keep getting "undefined: sin" compilation errors.

A:	Make sure you're actually linking with the math library.  For
	instance, due to a longstanding bug in Unix and Linux systems,
	you usually need to use an explicit -lm flag, at the *end* of
	the command line, when compiling/linking.  See also questions
	13.25, 13.26, and 14.2.

14.4a:	My floating-point calculations are acting strangely and giving
	me different answers on different machines.

A:	First, see question 14.2 above.

	If the problem isn't that simple, recall that digital computers
	usually use floating-point formats which provide a close but by
	no means exact simulation of real number arithmetic.  Underflow,
	cumulative precision loss, and other anomalies are often
	troublesome.

	Don't assume that floating-point results will be exact, and
	especially don't assume that floating-point values can be
	compared for equality.  (Don't throw haphazard "fuzz factors"
	in, either; see question 14.5.)

	These problems are no worse for C than they are for any other
	computer language.  Certain aspects of floating-point are
	usually defined as "however the processor does them" (see also
	question 11.34), otherwise a compiler for a machine without the
	"right" model would have to do prohibitively expensive
	emulations.

	This article cannot begin to list the pitfalls associated with,
	and workarounds appropriate for, floating-point work.  A good
	numerical programming text should cover the basics; see also the
	references below.

	References: Kernighan and Plauger, _The Elements of Programming
	Style_ Sec. 6 pp. 115-8; Knuth, Volume 2 chapter 4; David
	Goldberg, "What Every Computer Scientist Should Know about
	Floating-Point Arithmetic".

14.5:	What's a good way to check for "close enough" floating-point
	equality?

A:	Since the absolute accuracy of floating point values varies, by
	definition, with their magnitude, the best way of comparing two
	floating point values is to use an accuracy threshold which is
	relative to the magnitude of the numbers being compared.  Rather
	than

		double a, b;
		...
		if(a == b)	/* WRONG */

	use something like

		#include 

		if(fabs(a - b) <= epsilon * fabs(a))

	where epsilon is a value chosen to set the degree of "closeness"
	(and where you know that a will not be zero).

	References: Knuth Sec. 4.2.2 pp. 217-8.

14.6:	How do I round numbers?

A:	The simplest and most straightforward way is with code like

		(int)(x + 0.5)

	This technique won't work properly for negative numbers,
	though (for which you could use something like
	(int)(x < 0 ? x - 0.5 : x + 0.5)).

14.7:	Why doesn't C have an exponentiation operator?

A:	Because few processors have an exponentiation instruction.
	C has a pow() function, declared in , although explicit
	multiplication is usually better for small positive integral
	exponents.

	References: ISO Sec. 7.5.5.1; H&S Sec. 17.6 p. 393.

14.8:	The predefined constant M_PI seems to be missing from my
	machine's copy of .

A:	That constant (which is apparently supposed to be the value of
	pi, accurate to the machine's precision), is not standard.  If
	you need pi, you'll have to define it yourself, or compute it
	with 4*atan(1.0) or acos(-1.0).

	References: PCS Sec. 13 p. 237.

14.9:	How do I test for IEEE NaN and other special values?

A:	Many systems with high-quality IEEE floating-point
	implementations provide facilities (e.g. predefined constants,
	and functions like isnan(), either as nonstandard extensions in
	 or perhaps in  or ) to deal with these
	values cleanly, and work is being done to formally standardize
	such facilities.  A crude but usually effective test for NaN is
	exemplified by

		#define isnan(x) ((x) != (x))

	although non-IEEE-aware compilers may optimize the test away.

	C99 provides isnan(), fpclassify(), and several other
	classification routines.

	Another possibility is to format the value in question using
	sprintf(): on many systems it generates strings like "NaN" and
	"Inf" which you could compare for in a pinch.

	See also question 19.39.

	References: C9X Sec. 7.7.3.

14.11:	What's a good way to implement complex numbers in C?

A:	It is straightforward to define a simple structure and some
	arithmetic functions to manipulate them.  C99 supports complex
	as a standard type.  See also questions 2.10 and 14.12.

	References: C9X Sec. 6.1.2.5, Sec. 7.8.

14.12:	I'm looking for some code to do:
		Fast Fourier Transforms (FFT's)
		matrix arithmetic (multiplication, inversion, etc.)
		complex arithmetic

A:	Ajay Shah has prepared a nice index of free numerical
	software which has been archived pretty widely; one URL
	is ftp://ftp.math.psu.edu/pub/FAQ/numcomp-free-c .
	See also questions 18.9b, 18.13, 18.15c, and 18.16.

14.13:	I'm having trouble with a Turbo C program which crashes and says
	something like "floating point formats not linked."

A:	Some compilers for small machines, including Turbo C (and
	Ritchie's original PDP-11 compiler), leave out certain floating
	point support if it looks like it will not be needed.  In
	particular, the non-floating-point versions of printf() and
	scanf() save space by not including code to handle %e, %f,
	and %g.  It happens that Borland's heuristics for determining
	whether the program uses floating point are insufficient,
	and the programmer must sometimes insert a dummy call to a
	floating-point library function (such as sqrt(); any will
	do) to force loading of floating-point support.  (See the
	comp.os.msdos.programmer FAQ list for more information.)


Section 15. Variable-Length Argument Lists

15.1:	I heard that you have to #include  before calling
	printf().  Why?

A:	So that a proper prototype for printf() will be in scope.

	A compiler may use a different calling sequence for functions
	which accept variable-length argument lists.  (It might do so if
	calls using variable-length argument lists were less efficient
	than those using fixed-length.)  Therefore, a prototype
	(indicating, using the ellipsis notation "...", that the
	argument list is of variable length) must be in scope whenever a
	varargs function is called, so that the compiler knows to use
	the varargs calling mechanism.

	References: ISO Sec. 6.3.2.2, Sec. 7.1.7; Rationale
	Sec. 3.3.2.2, Sec. 4.1.6; H&S Sec. 9.2.4 pp. 268-9, Sec. 9.6 pp.
	275-6.

15.2:	How can %f be used for both float and double arguments in
	printf()?  Aren't they different types?

A:	In the variable-length part of a variable-length argument list,
	the "default argument promotions" apply: types char and
	short int are promoted to int, and float is promoted to double.
	(These are the same promotions that apply to function calls
	without a prototype in scope, also known as "old style" function
	calls; see question 11.3.)  Therefore, printf's %f format always
	sees a double.  (Similarly, %c always sees an int, as does %hd.)
	See also questions 12.9 and 12.13.

	References: ISO Sec. 6.3.2.2; H&S Sec. 6.3.5 p. 177, Sec. 9.4
	pp. 272-3.

15.3:	I had a frustrating problem which turned out to be caused by the
	line

		printf("%d", n);

	where n was actually a long int.  I thought that ANSI function
	prototypes were supposed to guard against argument type
	mismatches like this.

A:	When a function accepts a variable number of arguments, its
	prototype does not (and cannot) provide any information about
	the number and types of those variable arguments.  Therefore,
	the usual protections do *not* apply in the variable-length part
	of variable-length argument lists: the compiler cannot perform
	implicit conversions or (in general) warn about mismatches.

	See also questions 5.2, 11.3, 12.9, and 15.2.

15.4:	How can I write a function that takes a variable number of
	arguments?

A:	Use the facilities of the  header.

	Here is a function which concatenates an arbitrary number of
	strings into malloc'ed memory:

		#include 		/* for malloc, NULL, size_t */
		#include 		/* for va_ stuff */
		#include 		/* for strcat et al. */

		char *vstrcat(const char *first, ...)
		{
			size_t len;
			char *retbuf;
			va_list argp;
			char *p;

			if(first == NULL)
				return NULL;

			len = strlen(first);

			va_start(argp, first);

			while((p = va_arg(argp, char *)) != NULL)
				len += strlen(p);

			va_end(argp);

			retbuf = malloc(len + 1);	/* +1 for trailing /0 */

			if(retbuf == NULL)
				return NULL;		/* error */

			(void)strcpy(retbuf, first);

			va_start(argp, first);		/* restart; 2nd scan */

			while((p = va_arg(argp, char *)) != NULL)
				(void)strcat(retbuf, p);

			va_end(argp);

			return retbuf;
		}

	Usage is something like

		char *str = vstrcat("Hello, ", "world!", (char *)NULL);

	Note the cast on the last argument; see questions 5.2 and 15.3.
	(Also note that the caller must free the returned, malloc'ed
	storage.)

	References: K&R2 Sec. 7.3 p. 155, Sec. B7 p. 254; ISO Sec. 7.8;
	Rationale Sec. 4.8; H&S Sec. 11.4 pp. 296-9; CT&P Sec. A.3 pp.
	139-141; PCS Sec. 11 pp. 184-5, Sec. 13 p. 242.

15.5:	How can I write a function that takes a format string and a
	variable number of arguments, like printf(), and passes them to
	printf() to do most of the work?

A:	Use vprintf(), vfprintf(), or vsprintf().

	Here is an error() function which prints an error message,
	preceded by the string "error: " and terminated with a newline:

		#include 
		#include 

		void error(const char *fmt, ...)
		{
			va_list argp;
			fprintf(stderr, "error: ");
			va_start(argp, fmt);
			vfprintf(stderr, fmt, argp);
			va_end(argp);
			fprintf(stderr, "/n");
		}

	References: K&R2 Sec. 8.3 p. 174, Sec. B1.2 p. 245; ISO
	Secs. 7.9.6.7,7.9.6.8,7.9.6.9; H&S Sec. 15.12 pp. 379-80; PCS
	Sec. 11 pp. 186-7.

15.6:	How can I write a function analogous to scanf(), that calls
	scanf() to do most of the work?

A:	C99 (but *not* any earlier C Standard) supports vscanf(),
	vfscanf(), and vsscanf().

	References: C9X Secs. 7.3.6.12-14.

15.8:	How can I discover how many arguments a function was actually
	called with?

A:	This information is not available to a portable program.  Some
	old systems provided a nonstandard nargs() function, but its use
	was always questionable, since it typically returned the number
	of words passed, not the number of arguments.  (Structures, long
	ints, and floating point values are usually passed as several
	words.)

	Any function which takes a variable number of arguments must be
	able to determine *from the arguments themselves* how many of
	them there are.  printf-like functions do this by looking for
	formatting specifiers (%d and the like) in the format string
	(which is why these functions fail badly if the format string
	does not match the argument list).  Another common technique,
	applicable when the arguments are all of the same type, is to
	use a sentinel value (often 0, -1, or an appropriately-cast null
	pointer) at the end of the list (see the execl() and vstrcat()
	examples in questions 5.2 and 15.4).  Finally, if the types are
	predictable, you can pass an explicit count of the number of
	variable arguments (although it's usually a nuisance for the
	caller to supply).

	References: PCS Sec. 11 pp. 167-8.

15.9:	My compiler isn't letting me declare a function

		int f(...)
		{
		}

	i.e. with no fixed arguments.

A:	Standard C requires at least one fixed argument, in part so that
	you can hand it to va_start().  See also question 15.10.

	References: ISO Sec. 6.5.4, Sec. 6.5.4.3, Sec. 7.8.1.1; H&S
	Sec. 9.2 p. 263.

15.10:	I have a varargs function which accepts a float parameter.  Why
	isn't

		va_arg(argp, float)

	working?

A:	In the variable-length part of variable-length argument lists,
	the old "default argument promotions" apply: arguments of type
	float are always promoted (widened) to type double, and types
	char and short int are promoted to int.  Therefore, it is never
	correct to invoke va_arg(argp, float); instead you should always
	use va_arg(argp, double).  Similarly, use va_arg(argp, int) to
	retrieve arguments which were originally char, short, or int.
	(For analogous reasons, the last "fixed" argument, as handed to
	va_start(), should not be widenable, either.)  See also
	questions 11.3 and 15.2.

	References: ISO Sec. 6.3.2.2; Rationale Sec. 4.8.1.2; H&S
	Sec. 11.4 p. 297.

15.11:	I can't get va_arg() to pull in an argument of type pointer-to-
	function.

A:	The type-rewriting games which the va_arg() macro typically
	plays are stymied by overly-complicated types such as pointer-
	to-function.  If you use a typedef for the function pointer
	type, however, all will be well.  See also question 1.21.

	References: ISO Sec. 7.8.1.2; Rationale Sec. 4.8.1.2.

15.12:	How can I write a function which takes a variable number of
	arguments and passes them to some other function (which takes a
	variable number of arguments)?

A:	In general, you cannot.  Ideally, you should provide a version
	of that other function which accepts a va_list pointer
	(analogous to vfprintf(); see question 15.5 above).  If the
	arguments must be passed directly as actual arguments, or if you
	do not have the option of rewriting the second function to
	accept a va_list (in other words, if the second, called function
	must accept a variable number of arguments, not a va_list), no
	portable solution is possible.  (The problem could perhaps be
	solved by resorting to machine-specific assembly language; see
	also question 15.13 below.)

15.13:	How can I call a function with an argument list built up at run
	time?

A:	There is no guaranteed or portable way to do this.  If you're
	curious, ask this list's editor, who has a few wacky ideas you
	could try...

	Instead of an actual argument list, you might consider passing
	an array of generic (void *) pointers.  The called function can
	then step through the array, much like main() might step through
	argv.  (Obviously this works only if you have control over all
	the called functions.)

	(See also question 19.36.)


Section 16. Strange Problems

16.1b:	I'm getting baffling syntax errors which make no sense at all,
	and it seems like large chunks of my program aren't being
	compiled.

A:	Check for unclosed comments, mismatched #if/#ifdef/#ifndef/
	#else/#endif directives, and perhaps unclosed quotes; remember
	to check header files, too.  (See also questions 2.18, 10.9, and
	11.29.)

16.1c:	Why isn't my procedure call working?  The compiler seems to skip
	right over it.

A:	Does the code look like this?

		myprocedure;

	C has only functions, and function calls always require
	parenthesized argument lists, even if empty.  Use

		myprocedure();

16.3:	This program crashes before it even runs!  (When single-stepping
	with a debugger, it dies before the first statement in main().)

A:	You probably have one or more very large (kilobyte or more)
	local arrays.  Many systems have fixed-size stacks, and even
	those which perform dynamic stack allocation automatically
	(e.g. Unix) can be confused when the stack tries to grow by a
	huge chunk all at once.  It is often better to declare large
	arrays with static duration (unless of course you need a fresh
	set with each recursive call, in which case you could
	dynamically allocate them with malloc(); see also question
	1.31).

	(See also questions 11.12b, 16.4, 16.5, and 18.4.)

16.4:	I have a program that seems to run correctly, but it crashes as
	it's exiting, *after* the last statement in main().  What could
	be causing this?

A:	Look for a misdeclared main() (see questions 2.18, 10.9, 11.12b,
	and 11.14a), or local buffers passed to setbuf() or setvbuf(),
	or problems in cleanup functions registered by atexit().
	See also questions 7.5a and 11.16.

	References: CT&P Sec. 5.3 pp. 72-3.

16.5:	This program runs perfectly on one machine, but I get weird
	results on another.  Stranger still, adding or removing a
	debugging printout changes the symptoms...

A:	Lots of things could be going wrong; here are a few of the more
	common things to check:

		uninitialized local variables (see also question 7.1)

		integer overflow, especially on 16-bit machines,
		especially of an intermediate result when doing things
		like a * b / c (see also question 3.14)

		undefined evaluation order (see questions 3.1 through 3.4)

		omitted declaration of external functions, especially
		those which return something other than int, or have
		"narrow" or variable arguments (see questions 1.25,
		11.3, 14.2, and 15.1)

		dereferenced null pointers (see section 5)

		improper malloc/free use: assuming malloc'ed memory
		contains 0, assuming freed storage persists, freeing
		something twice, corrupting the malloc arena (see also
		questions 7.19 and 7.20)

		pointer problems in general (see also question 16.8)

		mismatch between printf() format and arguments, especially
		trying to print long ints using %d (see question 12.9)

		trying to allocate more memory than an unsigned int can
		count, especially on machines with limited memory (see
		also questions 7.16 and 19.23)

		array bounds problems, especially of small, temporary
		buffers, perhaps used for constructing strings with
		sprintf() (see also questions 7.1 and 12.21)

		invalid assumptions about the mapping of typedefs,
		especially size_t

		floating point problems (see questions 14.1 and 14.4a)

		anything you thought was a clever exploitation of the way
		you believe code is generated for your specific system

	Proper use of function prototypes can catch several of these
	problems; lint would catch several more.  See also questions
	16.3, 16.4, and 18.4.

16.6:	Why does this code:

		char *p = "hello, world!";
		p[0] = 'H';

	crash?

A:	String literals are not necessarily modifiable, except (in
	effect) when they are used as array initializers.  Try

		char a[] = "hello, world!";

	See also question 1.32.

	References: ISO Sec. 6.1.4; H&S Sec. 2.7.4 pp. 31-2.

16.8:	What do "Segmentation violation", "Bus error", and "General
	protection fault" mean?

A:	These generally mean that your program tried to access memory it
	shouldn't have, invariably as a result of stack corruption or
	improper pointer use.  Likely causes are overflow of local
	("automatic," stack-allocated) arrays; inadvertent use of null
	pointers (see also questions 5.2 and 5.20) or uninitialized,
	misaligned, or otherwise improperly allocated pointers (see
	questions 7.1 and 7.2); corruption of the malloc arena (see
	question 7.19); and mismatched function arguments, especially
	involving pointers; two possibilities are scanf() (see question
	12.12) and fprintf() (make sure it receives its first FILE *
	argument).

	See also questions 16.3 and 16.4.


Section 17. Style

17.1:	What's the best style for code layout in C?

A:	K&R, while providing the example most often copied, also supply
	a good excuse for disregarding it:

		The position of braces is less important,
		although people hold passionate beliefs.
		We have chosen one of several popular styles.
		Pick a style that suits you, then use it
		consistently.

	It is more important that the layout chosen be consistent (with
	itself, and with nearby or common code) than that it be
	"perfect."  If your coding environment (i.e. local custom or
	company policy) does not suggest a style, and you don't feel
	like inventing your own, just copy K&R.  (The tradeoffs between
	various indenting and brace placement options can be
	exhaustively and minutely examined, but don't warrant repetition
	here.  See also the Indian Hill Style Guide.)

	The elusive quality of "good style" involves much more than mere
	code layout details; don't spend time on formatting to the
	exclusion of more substantive code quality issues.

	See also question 10.6.

	References: K&R1 Sec. 1.2 p. 10; K&R2 Sec. 1.2 p. 10.

17.3:	Here's a neat trick for checking whether two strings are equal:

		if(!strcmp(s1, s2))

	Is this good style?

A:	It is not particularly good style, although it is a popular
	idiom.  The test succeeds if the two strings are equal, but the
	use of ! ("not") suggests that it tests for inequality.

	Another option is to use a macro:

		#define Streq(s1, s2) (strcmp((s1), (s2)) == 0)

	See also question 17.10.

17.4:	Why do some people write if(0 == x) instead of if(x == 0)?

A:	It's a trick to guard against the common error of writing

		if(x = 0)

	If you're in the habit of writing the constant before the ==,
	the compiler will complain if you accidentally type

		if(0 = x)

	Evidently it can be easier for some people to remember to
	reverse the test than to remember to type the doubled = sign.
	(Of course, the trick only helps when comparing to a constant.)

	References: H&S Sec. 7.6.5 pp. 209-10.

17.4b:	I've seen function declarations that look like this:

		extern int func __((int, int));

	What are those extra parentheses and underscores for?

A:	They're part of a trick which allows the prototype part of the
	function declaration to be turned off for a pre-ANSI compiler.
	Somewhere else is a conditional definition of the __ macro like
	this:

		#ifdef __STDC__
		#define __(proto) proto
		#else
		#define __(proto) ()
		#endif

	The extra parentheses in the invocation

		extern int func __((int, int));

	are required so that the entire prototype list (perhaps
	containing many commas) is treated as the single argument
	expected by the macro.

17.5:	I came across some code that puts a (void) cast before each call
	to printf().  Why?

A:	printf() does return a value, though few programs bother to
	check the return values from each call.  Since some compilers
	(and lint) will warn about discarded return values, an explicit
	cast to (void) is a way of saying "Yes, I've decided to ignore
	the return value from this call, but please continue to warn me
	about other (perhaps inadvertently) ignored return values."
	It's also common to use void casts on calls to strcpy() and
	strcat(), since the return value is never surprising.

	References: K&R2 Sec. A6.7 p. 199; Rationale Sec. 3.3.4; H&S
	Sec. 6.2.9 p. 172, Sec. 7.13 pp. 229-30.

17.8:	What is "Hungarian Notation"?  Is it worthwhile?

A:	Hungarian Notation is a naming convention, invented by Charles
	Simonyi, which encodes information about a variable's type (and
	perhaps its intended use) in its name.  It is well-loved in some
	circles and roundly castigated in others.  Its chief advantage
	is that it makes a variable's type or intended use obvious from
	its name; its chief disadvantage is that type information is not
	necessarily a worthwhile thing to carry around in the name of a
	variable.

	References: Simonyi and Heller, "The Hungarian Revolution" .

17.9:	Where can I get the "Indian Hill Style Guide" and other coding
	standards?

A:	Various documents are available for anonymous ftp from:

		Site:			File or directory:

		ftp.cs.washington.edu	pub/cstyle.tar.Z
					(the updated Indian Hill guide)

		ftp.cs.toronto.edu	doc/programming
					(including Henry Spencer's
					"10 Commandments for C Programmers")

		ftp.cs.umd.edu		pub/style-guide

	You may also be interested in the books _The Elements of
	Programming Style_, _Plum Hall Programming Guidelines_, and _C
	Style: Standards and Guidelines_; see the Bibliography.

	See also question 18.9.

17.10:	Some people say that goto's are evil and that I should never use
	them.  Isn't that a bit extreme?

A:	Programming style, like writing style, is somewhat of an art and
	cannot be codified by inflexible rules, although discussions
	about style often seem to center exclusively around such rules.

	In the case of the goto statement, it has long been observed
	that unfettered use of goto's quickly leads to unmaintainable
	spaghetti code.  However, a simple, unthinking ban on the goto
	statement does not necessarily lead immediately to beautiful
	programming: an unstructured programmer is just as capable of
	constructing a Byzantine tangle without using any goto's
	(perhaps substituting oddly-nested loops and Boolean control
	variables, instead).

	Most observations or "rules" about programming style usually
	work better as guidelines than rules, and work much better if
	programmers understand what the guidelines are trying to
	accomplish.  Blindly avoiding certain constructs or following
	rules without understanding them can lead to just as many
	problems as the rules were supposed to avert.

	Furthermore, many opinions on programming style are just that:
	opinions.  It's usually futile to get dragged into "style wars,"
	because on certain issues (such as those referred to in
	questions 5.3, 5.9, 9.2, and 10.7), opponents can never seem to
	agree, or agree to disagree, or stop arguing.


Section 18. Tools and Resources

[NOTE: Much of the information in this section is fairly old and may be
out-of-date, especially the URLs of various allegedly publicly-available
packages.  Caveat lector.]

18.1:	I need:			     A:	Look for programs (see also
					question 18.16) named:

	a C cross-reference		cflow, cxref, calls, cscope,
	generator			xscope, or ixfw

	a C beautifier/pretty-		cb, indent, GNU indent, or
	printer				vgrind

	a revision control or		CVS, RCS, or SCCS
	configuration management
	tool

	a C source obfuscator		obfus, shroud, or opqcp
	(shrouder)

	a "make" dependency		makedepend, or try cc -M or
	generator			cpp -M

	tools to compute code		ccount, Metre, lcount, or csize;
	metrics				there is also a package sold by
					McCabe and Associates

	a C lines-of-source		this can be done very crudely
	counter				with the standard Unix utility
					wc, and somewhat better with
					grep -c ";"

	a C declaration aid		check volume 14 of
	(cdecl)                         comp.sources.unix (see question
					18.16) and K&R2

	a prototype generator		see question 11.31

	a tool to track down		see question 18.2
	malloc problems

	a "selective" C			see question 10.18
	preprocessor

	language translation		see questions 11.31 and 20.26
	tools

	C verifiers (lint)		see question 18.7

	a C compiler!			see question 18.3

	(This list of tools is by no means complete; if you know of
	tools not mentioned, you're welcome to contact this list's
	maintainer.)

	Other lists of tools, and discussion about them, can be found in
	the Usenet newsgroups comp.compilers and comp.software-eng.

	See also questions 18.3 and 18.16.

18.2:	How can I track down these pesky malloc problems?

A:	A number of debugging packages exist to help track down malloc
	problems; one popular one is Conor P. Cahill's "dbmalloc",
	posted to comp.sources.misc in 1992, volume 32.  Others are
	"leak", available in volume 27 of the comp.sources.unix
	archives; JMalloc.c and JMalloc.h in the "Snippets" collection;
	MEMDEBUG from ftp.crpht.lu in pub/sources/memdebug ; and
	Electric Fence.  See also question 18.16.

	A number of commercial debugging tools exist, and can be
	invaluable in tracking down malloc-related and other stubborn
	problems:

		CodeCenter (formerly Saber-C) from Centerline Software
		(http://www.centerline.com/).

		Insight (now Insure?), from ParaSoft Corporation
		(http://www.parasoft.com/).

		Purify, from Rational Software (http://www-
		306.ibm.com/software/rational/, formerly Pure Software,
		now part of IBM).

		ZeroFault, from The ZeroFault Group,
		http://www.zerofault.com/.

18.3:	What's a free or cheap C compiler I can use?

A:	A popular and high-quality free C compiler is the FSF's GNU C
	compiler, or gcc; see the gcc home page at http://gcc.gnu.org/.
	An MS-DOS port, djgpp, is also available; see the djgpp home
	page at http://www.delorie.com/djgpp/.  As far as I know, there
	are versions of gcc for Macs and Windows machines, too.

	Another popular compiler is lcc, described at
	http://www.cs.virginia.edu/~lcc-win32/ and
	http://www.cs.princeton.edu/software/lcc/.

	A very inexpensive MS-DOS compiler is Power C from Mix Software,
	1132 Commerce Drive, Richardson, TX 75801, USA, 214-783-6001.

	A shareware MS-DOS C compiler is available from
	ftp.hitech.com.au/hitech/pacific.  Registration is optional for
	non-commercial use.

	Archives associated with the comp.compilers newsgroup contain a
	great deal of information about available compilers,
	interpreters, grammars, etc. (for many languages).  The
	comp.compilers archives at http://compilers.iecc.com/ include an
	FAQ list and a catalog of free compilers.

	See also question 18.16.

18.4:	I just typed in this program, and it's acting strangely.  Can
	you see anything wrong with it?

A:	See if you can run lint first (perhaps with the -a, -c, -h, -p
	or other options).  Many C compilers are really only half-
	compilers, electing not to diagnose numerous source code
	difficulties which would not actively preclude code generation.

	See also questions 16.5, 16.8, and 18.7.

	References: Ian Darwin, _Checking C Programs with lint_ .

18.7:	Where can I get an ANSI-compatible lint?

A:	Products called PC-Lint and FlexeLint are available from Gimpel
	Software at http://www.gimpel.com/.

	The Unix System V release 4 lint is ANSI-compatible, and is
	available separately (bundled with other C tools) from UNIX
	Support Labs or from System V resellers.

	Another ANSI-compatible lint (which can also perform higher-
	level formal verification) is Splint (formerly lclint) at
	http://lclint.cs.virginia.edu/.

	In the absence of lint, many modern compilers do attempt to
	diagnose almost as many problems as lint does.  (Many netters
	recommend gcc -Wall -pedantic .)

18.8:	Don't ANSI function prototypes render lint obsolete?

A:	Not really.  First of all, prototypes work only if they are
	present and correct; an inadvertently incorrect prototype is
	worse than useless.  Secondly, lint checks consistency across
	multiple source files, and checks data declarations as well as
	functions.  Finally, an independent program like lint will
	probably always be more scrupulous at enforcing compatible,
	portable coding practices than will any particular,
	implementation-specific, feature- and extension-laden compiler.

	If you do want to use function prototypes instead of lint for
	cross-file consistency checking, make sure that you set the
	prototypes up correctly in header files.  See questions 1.7 and
	10.6.

18.9:	Are there any C tutorials or other resources on the net?

A:	There are several of them:

	Tom Torfs has a nice tutorial at http://cprog.tomsweb.net .

	"Notes for C programmers," by Christopher Sawtell, are
	available by ftp from svr-ftp.eng.cam.ac.uk in
	misc/sawtell_C.shar and garbo.uwasa.fi in pc/c-lang/c-
	lesson.zip, or on the web at
	http://www.fi.uib.no/Fysisk/Teori/KURS/OTHER/newzealand.html .

	Tim Love's "C for Programmers" is available by ftp from svr-
	ftp.eng.cam.ac.uk in the misc directory.  An html version is at
	http://www-h.eng.cam.ac.uk/help/tpl/languages/C/teaching_C/
	teaching_C.html .

	The Coronado Enterprises C tutorials are available on Simtel
	mirrors in pub/msdos/c or on the web at
	http://www.coronadoenterprises.com/tutorials/c/index.html .

	There is a web-based course by Steve Holmes at
	http://www.strath.ac.uk/IT/Docs/Ccourse/ .

	Martin Brown has C course material on the web at
	http://www-isis.ecs.soton.ac.uk/computing/c/Welcome.html .

	On some Unix machines you can try typing "learn c" at the shell
	prompt (but the lessons may be quite dated).

	Finally, the author of this FAQ list once taught a couple of
	C classes and has placed their notes on the web; they are at
	http://www.eskimo.com/~scs/cclass/cclass.html .

	[Disclaimer: I have not reviewed many of these tutorials, and
	I gather that they tend to contain errors.  With the exception
	of the one with my name on it, I can't vouch for any of them.
	Also, this sort of information rapidly becomes out-of-date;
	these addresses may not work by the time you read this and
	try them.]

	Several of these tutorials, plus a great deal of other
	information about C, are accessible via the web at
	http://www.lysator.liu.se/c/index.html .

	Vinit Carpenter maintains a list of resources for learning C and
	C++; it is posted to comp.lang.c and comp.lang.c++, and archived
	where this FAQ list is (see question 20.40), or on the web at
	http://www.cyberdiem.com/vin/learn.html .

	See also questions 18.9b, 18.10, and 18.15c.

18.9b:	Where can I find some good code examples to study and learn
	from?

A:	Here are a couple of links to explore:

		ftp://garbo.uwasa.fi/pc/c-lang/00index.txt

		http://www.eskimo.com/~scs/src/

	(Beware, though, that there is all too much truly bletcherous
	code out there, too.  Don't "learn" from bad code that it's the
	best anyone can do; you can do better.)  See also questions
	18.9, 18.13, 18.15c, and 18.16.

18.10:	What's a good book for learning C?  What about advanced books
	and references?

A:	There are far too many books on C to list here; it's impossible
	to rate them all.  Many people believe that the best one was
	also the first: _The C Programming Language_, by Kernighan and
	Ritchie ("K&R," now in its second edition).  Opinions vary on
	K&R's suitability as an initial programming text: many of us did
	learn C from it, and learned it well; some, however, feel that
	it is a bit too clinical as a first tutorial for those without
	much programming background.  Several sets of annotations and
	errata are available on the net, see e.g.
	http://www.csd.uwo.ca/~jamie/.Refs/.Footnotes/C-annotes.html ,
	http://www.eskimo.com/~scs/cclass/cclass.html , and
	http://cm.bell-labs.com/cm/cs/cbook/2ediffs.html .

	Many comp.lang.c regulars recommend _C: A Modern Approach_,
	by K.N. King.

	An excellent reference manual is _C: A Reference Manual_, by
	Samuel P. Harbison and Guy L. Steele, now in its fourth edition.

	Though not suitable for learning C from scratch, this FAQ list
	has been published in book form; see the Bibliography.

	The Association of C and C++ Users (ACCU) maintains a
	comprehensive set of bibliographic reviews of C/C++ titles at
	http://www.accu.org/bookreviews/public/.

	See also question 18.9 above.

18.13:	Where can I find the sources of the standard C libraries?

A:	The GNU project has a complete implementation at
	http://www.gnu.org/software/libc/.  Another source (though not
	public domain) is _The Standard C Library_, by P.J. Plauger (see
	the Bibliography).  See also questions 18.9b, 18.15c, and 18.16.

18.13b:	Is there an on-line C reference manual?

A:	Two possibilities are
	http://www.cs.man.ac.uk/standard_c/_index.html and
	http://www.dinkumware.com/htm_cl/index.html .

18.13c:	Where can I get a copy of the ANSI/ISO C Standard?

A:	See question 11.2.

18.14:	I need code to parse and evaluate expressions.

A:	Two available packages are "defunc," posted to comp.sources.misc
	in December, 1993 (V41 i32,33), to alt.sources in January, 1994,
	and available from sunsite.unc.edu in
	pub/packages/development/libraries/defunc-1.3.tar.Z, and
	"parse," at lamont.ldgo.columbia.edu.  Other options include the
	S-Lang interpreter, available via anonymous ftp from
	amy.tch.harvard.edu in pub/slang, and the shareware Cmm ("C-
	minus-minus" or "C minus the hard stuff").  See also questions
	18.16 and 20.6.

	There is also some parsing/evaluation code in _Software
	Solutions in C_ (chapter 12, pp. 235-55).

18.15:	Where can I get a BNF or YACC grammar for C?

A:	The definitive grammar is of course the one in the ANSI
	standard; see question 11.2.  Another grammar by Jim Roskind
	is available at ftp.eskimo.com in u/s/scs/roskind_grammar.Z .
	A fleshed-out, working instance of the ANSI C90 grammar
	(due to Jeff Lee) is on ftp.uu.net (see question 18.16) in
	usenet/net.sources/ansi.c.grammar.Z (including a companion
	lexer).  The FSF's GNU C compiler contains a grammar, as does
	the appendix to K&R2.

	The comp.compilers archives contain more information about
	grammars; see question 18.3.

	References: K&R1 Sec. A18 pp. 214-219; K&R2 Sec. A13 pp.
	234-239; ISO Sec. B.2; H&S pp. 423-435 Appendix B.

18.15b:	Does anyone have a C compiler test suite I can use?

A:	Plum Hall (formerly in Cardiff, NJ; now in Hawaii) sells one;
	other packages are Ronald Guilmette's RoadTest(tm) Compiler Test
	Suites (ftp to netcom.com, pub/rfg/roadtest/announce.txt for
	information) and Nullstone's Automated Compiler Performance
	Analysis Tool (see http://www.nullstone.com).  The FSF's GNU C
	(gcc) distribution includes a c-torture-test which checks a
	number of common problems with compilers.  Kahan's paranoia
	test, found in netlib/paranoia on netlib.att.com, strenuously
	tests a C implementation's floating point capabilities.

18.15c:	Where are some collections of useful code fragments and
	examples?

A:	Bob Stout's popular "SNIPPETS" collection is available from
	ftp.brokersys.com in directory pub/snippets or on the web at
	http://www.brokersys.com/snippets/ .

	Lars Wirzenius's "publib" library is available from ftp.funet.fi
	in directory pub/languages/C/Publib/.

	See also questions 14.12, 18.9, 18.9b, 18.13, and 18.16.

18.15d:	I need code for performing multiple precision arithmetic.

A:	Some popular packages are the "quad" functions within the BSD
	Unix libc sources (ftp.uu.net, /systems/unix/bsd-sources/.../
	src/lib/libc/quad/*), the GNU MP library "libmp", the MIRACL
	package (see http://indigo.ie/~mscott/ ), the "calc" program by
	David Bell and Landon Curt Noll, and the old Unix libmp.a.
	See also questions 14.12 and 18.16.

	References: Schumacher, ed., _Software Solutions in C_ Sec. 17
	pp. 343-454.

18.16:	Where and how can I get copies of all these freely distributable
	programs?

A:	As the number of available programs, the number of publicly
	accessible archive sites, and the number of people trying to
	access them all grow, this question becomes both easier and more
	difficult to answer.

	There are a number of large, public-spirited archive sites out
	there, such as ftp.uu.net, archive.umich.edu, oak.oakland.edu,
	sumex-aim.stanford.edu, and wuarchive.wustl.edu, which have huge
	amounts of software and other information all freely available.
	For the FSF's GNU project, the central distribution site is
	prep.ai.mit.edu .  These well-known sites tend to be extremely
	busy and hard to reach, but there are also numerous "mirror"
	sites which try to spread the load around.

	On the connected Internet, the traditional way to retrieve files
	from an archive site is with anonymous ftp.  For those without
	ftp access, there are also several ftp-by-mail servers in
	operation.  More and more, the world-wide web (WWW) is being
	used to announce, index, and even transfer large data files.
	There are probably yet newer access methods, too.

	Those are some of the easy parts of the question to answer.  The
	hard part is in the details -- this article cannot begin to
	track or list all of the available archive sites or all of the
	various ways of accessing them.  If you have access to the net
	at all, you probably have access to more up-to-date information
	about active sites and useful access methods than this FAQ list
	does.

	The other easy-and-hard aspect of the question, of course, is
	simply *finding* which site has what you're looking for.  There
	is a tremendous amount of work going on in this area, and there
	are probably new indexing services springing up every day.  One
	of the first was "archie", and of course there are a number of
	high-profile commercial net indexing and searching services such
	as Alta Vista, Excite, and Yahoo.

	If you have access to Usenet, see the regular postings in the
	comp.sources.unix and comp.sources.misc newsgroups, which
	describe the archiving policies for those groups and how to
	access their archives, two of which are
	ftp://gatekeeper.dec.com/pub/usenet/comp.sources.unix/ and
	ftp://ftp.uu.net/usenet/comp.sources.unix/.  The comp.archives
	newsgroup contains numerous announcements of anonymous ftp
	availability of various items.  Finally, the newsgroup
	comp.sources.wanted is generally a more appropriate place to
	post queries for source availability, but check *its* FAQ list,
	"How to find sources," before posting there.

	See also questions 14.12, 18.9b, 18.13, and 18.15c.


Section 19. System Dependencies

19.1:	How can I read a single character from the keyboard without
	waiting for the RETURN key?  How can I stop characters from
	being echoed on the screen as they're typed?

A:	Alas, there is no standard or portable way to do these things in
	C.  Concepts such as screens and keyboards are not even
	mentioned in the Standard, which deals only with simple I/O
	"streams" of characters.

	At some level, interactive keyboard input is usually collected
	and presented to the requesting program a line at a time.  This
	gives the operating system a chance to support input line
	editing (backspace/delete/rubout, etc.) in a consistent way,
	without requiring that it be built into every program.  Only
	when the user is satisfied and presses the RETURN key (or
	equivalent) is the line made available to the calling program.
	Even if the calling program appears to be reading input a
	character at a time (with getchar() or the like), the first call
	blocks until the user has typed an entire line, at which point
	potentially many characters become available and many character
	requests (e.g. getchar() calls) are satisfied in quick
	succession.

	When a program wants to read each character immediately as it
	arrives, its course of action will depend on where in the input
	stream the line collection is happening and how it can be
	disabled.  Under some systems (e.g. MS-DOS, VMS in some modes),
	a program can use a different or modified set of OS-level input
	calls to bypass line-at-a-time input processing.  Under other
	systems (e.g. Unix, VMS in other modes), the part of the
	operating system responsible for serial input (often called the
	"terminal driver") must be placed in a mode which turns off
	line-at-a-time processing, after which all calls to the usual
	input routines (e.g. read(), getchar(), etc.) will return
	characters immediately.  Finally, a few systems (particularly
	older, batch-oriented mainframes) perform input processing in
	peripheral processors which cannot be told to do anything other
	than line-at-a-time input.

	Therefore, when you need to do character-at-a-time input (or
	disable keyboard echo, which is an analogous problem), you will
	have to use a technique specific to the system you're using,
	assuming it provides one.  Since comp.lang.c is oriented towards
	those topics that the C language has defined support for, you
	will usually get better answers to other questions by referring
	to a system-specific newsgroup such as comp.unix.questions or
	comp.os.msdos.programmer, and to the FAQ lists for these groups.
	Note that the answers may differ even across variants of
	otherwise similar systems (e.g. across different variants of
	Unix); bear in mind when answering system-specific questions
	that the answer that applies to your system may not apply to
	everyone else's.

	However, since these questions are frequently asked here, here
	are brief answers for some common situations.

	Some versions of curses have functions called cbreak(),
	noecho(), and getch() which do what you want.  If you're
	specifically trying to read a short password without echo, you
	might try getpass().  Under Unix, you can use ioctl() to play
	with the terminal driver modes (CBREAK or RAW under "classic"
	versions; ICANON, c_cc[VMIN] and c_cc[VTIME] under System V or
	POSIX systems; ECHO under all versions), or in a pinch, system()
	and the stty command.  (For more information, see  and
	tty(4) under classic versions,  and termio(4) under
	System V, or  and termios(4) under POSIX.)  Under
	MS-DOS, use getch() or getche(), or the corresponding BIOS
	interrupts.  Under VMS, try the Screen Management (SMG$)
	routines, or curses, or issue low-level $QIO's with the
	IO$_READVBLK function code (and perhaps IO$M_NOECHO, and others)
	to ask for one character at a time.  (It's also possible to set
	character-at-a-time or "pass through" modes in the VMS terminal
	driver.)  Under other operating systems, you're on your own.

	(As an aside, note that simply using setbuf() or setvbuf() to
	set stdin to unbuffered will *not* generally serve to allow
	character-at-a-time input.)

	If you're trying to write a portable program, a good approach is
	to define your own suite of three functions to (1) set the
	terminal driver or input system into character-at-a-time mode
	(if necessary), (2) get characters, and (3) return the terminal
	driver to its initial state when the program is finished.
	(Ideally, such a set of functions might be part of the C
	Standard, some day.)  The extended versions of this FAQ list
	(see question 20.40) contain examples of such functions for
	several popular systems.

	See also question 19.2.

	References: PCS Sec. 10 pp. 128-9, Sec. 10.1 pp. 130-1; POSIX
	Sec. 7.

19.2:	How can I find out if there are characters available for reading
	(and if so, how many)?  Alternatively, how can I do a read that
	will not block if there are no characters available?

A:	These, too, are entirely operating-system-specific.  Some
	versions of curses have a nodelay() function.  Depending on your
	system, you may also be able to use "nonblocking I/O", or a
	system call named "select" or "poll", or the FIONREAD ioctl, or
	c_cc[VTIME], or kbhit(), or rdchk(), or the O_NDELAY option to
	open() or fcntl().  See also question 19.1.

19.3:	How can I display a percentage-done indication that updates
	itself in place, or show one of those "twirling baton" progress
	indicators?

A:	These simple things, at least, you can do fairly portably.
	Printing the character '/r' will usually give you a carriage
	return without a line feed, so that you can overwrite the
	current line.  The character '/b' is a backspace, and will
	usually move the cursor one position to the left.  (But remember
	to call fflush(), too.)

	References: ISO Sec. 5.2.2.

19.4:	How can I clear the screen?
	How can I print text in color?
	How can I move the cursor to a specific x, y position?

A:	Such things depend on the terminal type (or display) you're
	using.  You will have to use a library such as termcap,
	terminfo, or curses, or some system-specific routines, to
	perform these operations.  On MS-DOS systems, two functions
	to look for are clrscr() and gotoxy().

	For clearing the screen, a halfway portable solution is to print
	a form-feed character ('/f'), which will cause some displays to
	clear.  Even more portable (albeit even more gunky) might be to
	print enough newlines to scroll everything away.  As a last
	resort, you could use system() (see question 19.27) to invoke
	an operating system clear-screen command.

	References: PCS Sec. 5.1.4 pp. 54-60, Sec. 5.1.5 pp. 60-62.

19.5:	How do I read the arrow keys?  What about function keys?

A:	Terminfo, some versions of termcap, and some versions of curses
	have support for these non-ASCII keys.  Typically, a special key
	sends a multicharacter sequence (usually beginning with ESC,
	'/033'); parsing these can be tricky.  (curses will do the
	parsing for you, if you call keypad() first.)

	Under MS-DOS, if you receive a character with value 0 (*not*
	'0'!) while reading the keyboard, it's a flag indicating that
	the next character read will be a code indicating a special key.
	See any DOS programming guide for lists of keyboard scan codes.
	(Very briefly: the up, left, right, and down arrow keys are 72,
	75, 77, and 80, and the function keys are 59 through 68.)

	References: PCS Sec. 5.1.4 pp. 56-7.

19.6:	How do I read the mouse?

A:	Consult your system documentation, or ask on an appropriate
	system-specific newsgroup (but check its FAQ list first).  Mouse
	handling is completely different under the X window system, MS-
	DOS, the Macintosh, and probably every other system.

	References: PCS Sec. 5.5 pp. 78-80.

19.7:	How can I do serial ("comm") port I/O?

A:	It's system-dependent.  Under Unix, you typically open, read,
	and write a device file in /dev, and use the facilities of the
	terminal driver to adjust its characteristics.  (See also
	questions 19.1 and 19.2.)  Under MS-DOS, you can use the
	predefined stream stdaux, or a special file like COM1, or some
	primitive BIOS interrupts, or (if you require decent
	performance) any number of interrupt-driven serial I/O packages.
	Several netters recommend the book _C Programmer's Guide to
	Serial Communications_, by Joe Campbell.

19.8:	How can I direct output to the printer?

A:	Under Unix, either use popen() (see question 19.30) to write to
	the lp or lpr program, or perhaps open a special file like
	/dev/lp.  Under MS-DOS, write to the (nonstandard) predefined
	stdio stream stdprn, or open the special files PRN or LPT1.
	Under some circumstances, another (and perhaps the only)
	possibility is to use a window manager's screen-capture
	function, and print the resulting bitmap.

	References: PCS Sec. 5.3 pp. 72-74.

19.9:	How do I send escape sequences to control a terminal or other
	device?

A:	If you can figure out how to send characters to the device at
	all (see question 19.8 above), it's easy enough to send escape
	sequences.  In ASCII, the ESC code is 033 (27 decimal), so code
	like

		fprintf(ofd, "/033[J");

	sends the sequence ESC [ J .

19.9b:	How can I access an I/O board directly?

A:	In general, there are two ways to do this: use system-specific
	functions such as "inport" and "outport" (if the device is
	accessed via an "I/O port"), or use contrived pointer variables
	to access "memory-mapped I/O" device locations.  See question
	19.25.

19.10:	How can I do graphics?

A:	Once upon a time, Unix had a fairly nice little set of device-
	independent plot functions described in plot(3) and plot(5).
	The GNU libplot library, written by Robert Maier, maintains
	the same spirit and supports many modern plot devices; see
	http://www.gnu.org/software/plotutils/plotutils.html .

	A modern, platform-independent graphics library (which also
	supports 3D graphics and animation) is OpenGL.  Other graphics
	standards which may be of interest are GKS and PHIGS.

	If you're programming for MS-DOS, you'll probably want to use
	libraries conforming to the VESA or BGI standards.

	If you're trying to talk to a particular plotter, making it draw
	is usually a matter of sending it the appropriate escape
	sequences; see also question 19.9.  The vendor may supply a C-
	callable library, or you may be able to find one on the net.

	If you're programming for a particular window system (Macintosh,
	X windows, Microsoft Windows), you will use its facilities; see
	the relevant documentation or newsgroup or FAQ list.

	References: PCS Sec. 5.4 pp. 75-77.

19.10b:	How can I display GIF and JPEG images?

A:	It will depend on your display environment, which may already
	provide these functions.  Reference JPEG software is at
	http://www.ijg.org/files/ .

19.11:	How can I check whether a file exists?  I want to warn the user
	if a requested input file is missing.

A:	It's surprisingly difficult to make this determination reliably
	and portably.  Any test you make can be invalidated if the file
	is created or deleted (i.e. by some other process) between the
	time you make the test and the time you try to open the file.

	Three possible test functions are stat(), access(), and fopen().
	(To make an approximate test using fopen(), just open for
	reading and close immediately, although failure does not
	necessarily indicate nonexistence.)  Of these, only fopen() is
	widely portable, and access(), where it exists, must be used
	carefully if the program uses the Unix set-UID feature.

	Rather than trying to predict in advance whether an operation
	such as opening a file will succeed, it's often better to try
	it, check the return value, and complain if it fails.
	(Obviously, this approach won't work if you're trying to avoid
	overwriting an existing file, unless you've got something like
	the O_EXCL file opening option available, which does just what
	you want in this case.)

	References: PCS Sec. 12 pp. 189,213; POSIX Sec. 5.3.1,
	Sec. 5.6.2, Sec. 5.6.3.

19.12:	How can I find out the size of a file, prior to reading it in?

A:	If the "size of a file" is the number of characters you'll be
	able to read from it in C, it can be difficult or impossible to
	determine this number exactly.

	Under Unix, the stat() call will give you an exact answer.
	Several other systems supply a Unix-like stat() which will give
	an approximate answer.  You can fseek() to the end and then use
	ftell(), or maybe try fstat(), but these tend to have the same
	sorts of problems: fstat() is not portable, and generally tells
	you the same thing stat() tells you; ftell() is not guaranteed
	to return a byte count except for binary files (but, strictly
	speaking, binary files don't necessarily support fseek to
	SEEK_END at all).  Some systems provide functions called
	filesize() or filelength(), but these are obviously not
	portable, either.

	Are you sure you have to determine the file's size in advance?
	Since the most accurate way of determining the size of a file as
	a C program will see it is to open the file and read it, perhaps
	you can rearrange the code to learn the size as it reads.

	References: ISO Sec. 7.9.9.4; H&S Sec. 15.5.1; PCS Sec. 12 p.
	213; POSIX Sec. 5.6.2.

19.12b:	How can I find the modification date and time of a file?

A:	The Unix and POSIX function is stat(), which several other
	systems supply as well.  (See also question 19.12.)

19.13:	How can a file be shortened in-place without completely clearing
	or rewriting it?

A:	BSD systems provide ftruncate(), several others supply chsize(),
	and a few may provide a (possibly undocumented) fcntl option
	F_FREESP.  Under MS-DOS, you can sometimes use write(fd, "", 0).
	However, there is no portable solution, nor a way to delete
	blocks at the beginning.  See also question 19.14.

19.14:	How can I insert or delete a line (or record) in the middle of a
	file?

A:	Short of rewriting the file, you probably can't.  The usual
	solution is simply to rewrite the file.  (Instead of deleting
	records, you might consider simply marking them as deleted, to
	avoid rewriting.)  Another possibility, of course, is to use a
	database instead of a flat file.  See also questions 12.30 and
	19.13.

19.15:	How can I recover the file name given an open stream or file
	descriptor?

A:	This problem is, in general, insoluble.  Under Unix, for
	instance, a scan of the entire disk (perhaps involving special
	permissions) would theoretically be required, and would fail if
	the descriptor were connected to a pipe or referred to a deleted
	file (and could give a misleading answer for a file with
	multiple links).  It is best to remember the names of files
	yourself as you open them (perhaps with a wrapper function
	around fopen()).

19.16:	How can I delete a file?

A:	The Standard C Library function is remove().  (This is therefore
	one of the few questions in this section for which the answer is
	*not* "It's system-dependent.")  On older, pre-ANSI Unix
	systems, remove() may not exist, in which case you can try
	unlink().

	References: K&R2 Sec. B1.1 p. 242; ISO Sec. 7.9.4.1; H&S
	Sec. 15.15 p. 382; PCS Sec. 12 pp. 208,220-221; POSIX
	Sec. 5.5.1, Sec. 8.2.4.

19.16b:	How do I copy files?

A:	Either use system() to invoke your operating system's copy
	utility (see question 19.27), or open the source and destination
	files (using fopen() or some lower-level file-opening system
	call), read characters or blocks of characters from the source
	file, and write them to the destination file.

	References: K&R Sec. 1, Sec. 7.

19.17:	Why can't I open a file by its explicit path?  The call

		fopen("c:/newdir/file.dat", "r")

	is failing.

A:	The file you actually requested -- with the characters /n and /f
	in its name -- probably doesn't exist, and isn't what you
	thought you were trying to open.

	In character constants and string literals, the backslash / is
	an escape character, giving special meaning to the character
	following it.  In order for literal backslashes in a pathname to
	be passed through to fopen() (or any other function) correctly,
	they have to be doubled, so that the first backslash in each
	pair quotes the second one:

		fopen("c://newdir//file.dat", "r")

	Alternatively, under MS-DOS, it turns out that forward slashes
	are also accepted as directory separators, so you could use

		fopen("c:/newdir/file.dat", "r")

	(Note, by the way, that header file names mentioned in
	preprocessor #include directives are *not* string literals, so
	you may not have to worry about backslashes there.)

19.17b:	fopen() isn't letting me open files like "$HOME/.profile" and
	"~/.myrcfile".

A:	Under Unix, at least, environment variables like $HOME, along
	with the home-directory notation involving the ~ character, are
	expanded by the shell, and there's no mechanism to perform these
	expansions automatically when you call fopen().

19.17c:	How can I suppress the dreaded MS-DOS "Abort, Retry, Ignore?"
	message?

A:	Among other things, you need to intercept the DOS Critical Error
	Interrupt, interrupt 24H.  See the comp.os.msdos.programmer FAQ
	list for more details.

19.18:	I'm getting an error, "Too many open files".  How can I increase
	the allowable number of simultaneously open files?

A:	There are typically at least two resource limitations on the
	number of simultaneously open files: the number of low-level
	"file descriptors" or "file handles" available in the operating
	system, and the number of FILE structures available in the stdio
	library.  Both must be sufficient.  Under MS-DOS systems, you
	can control the number of operating system file handles with a
	line in CONFIG.SYS.  Some compilers come with instructions (and
	perhaps a source file or two) for increasing the number of stdio
	FILE structures.

19.20:	How can I read a directory in a C program?

A:	See if you can use the opendir() and readdir() functions, which
	are part of the POSIX standard and are available on most Unix
	variants.  Implementations also exist for MS-DOS, VMS, and other
	systems.  (MS-DOS also has FINDFIRST and FINDNEXT routines which
	do essentially the same thing, and MS Windows has FindFirstFile
	and FindNextFile.)  readdir() returns just the file names; if
	you need more information about the file, try calling stat().
	To match filenames to some wildcard pattern, see question 13.7.

	References: K&R2 Sec. 8.6 pp. 179-184; PCS Sec. 13 pp. 230-1;
	POSIX Sec. 5.1; Schumacher, ed., _Software Solutions in C_
	Sec. 8.

19.22:	How can I find out how much memory is available?

A:	Your operating system may provide a routine which returns this
	information, but it's quite system-dependent.

19.23:	How can I allocate arrays or structures bigger than 64K?

A:	A reasonable computer ought to give you transparent access to
	all available memory.  If you're not so lucky, you'll either
	have to rethink your program's use of memory, or use various
	system-specific techniques.

	64K is (still) a pretty big chunk of memory.  No matter how much
	memory your computer has available, it's asking a lot to be able
	to allocate huge amounts of it contiguously.  (The C Standard
	does not guarantee that single objects can be 32K or larger,
	or 64K for C99.)  Often it's a good idea to use data
	structures which don't require that all memory be contiguous.
	For dynamically-allocated multidimensional arrays, you can
	use pointers to pointers, as illustrated in question 6.16.
	Instead of a large array of structures, you can use a linked
	list, or an array of pointers to structures.

	If you're using a PC-compatible (8086-based) system, and running
	up against a 64K or 640K limit, consider using "huge" memory
	model, or expanded or extended memory, or malloc variants such
	as halloc() or farmalloc(), or a 32-bit "flat" compiler (e.g.
	djgpp, see question 18.3), or some kind of a DOS extender, or
	another operating system.

	References: ISO Sec. 5.2.4.1; C9X Sec. 5.2.4.1.

19.24:	What does the error message "DGROUP data allocation exceeds 64K"
	mean, and what can I do about it?  I thought that using large
	model meant that I could use more than 64K of data!

A:	Even in large memory models, MS-DOS compilers apparently toss
	certain data (strings, some initialized global or static
	variables) into a default data segment, and it's this segment
	that is overflowing.  Either use less global data, or, if you're
	already limiting yourself to reasonable amounts (and if the
	problem is due to something like the number of strings), you may
	be able to coax the compiler into not using the default data
	segment for so much.  Some compilers place only "small" data
	objects in the default data segment, and give you a way (e.g.
	the /Gt option under Microsoft compilers) to configure the
	threshold for "small."

19.25:	How can I access memory (a memory-mapped device, or graphics
	memory) located at a certain address?

A:	Set a pointer, of the appropriate type, to the right number
	(using an explicit cast to assure the compiler that you really
	do intend this nonportable conversion):

		unsigned int *magicloc = (unsigned int *)0x12345678;

	Then, *magicloc refers to the location you want.  If the
	location is a memory-mapped I/O register, you will probably also
	want to use the volatile qualifier.  (Under MS-DOS, you may find
	a macro like MK_FP() handy for working with segments and offsets.)

	References: K&R1 Sec. A14.4 p. 210; K&R2 Sec. A6.6 p. 199; ISO
	Sec. 6.3.4; Rationale Sec. 3.3.4; H&S Sec. 6.2.7 pp. 171-2.

19.27:	How can I invoke another program (a standalone executable,
	or an operating system command) from within a C program?

A:	Use the library function system(), which does exactly that.
	Note that system's return value is at best the command's exit
	status (although even that is not guaranteed), and usually has
	nothing to do with the output of the command.  Note also that
	system() accepts a single string representing the command to be
	invoked; if you need to build up a complex command line, you can
	use sprintf().

	Depending on your operating system, you may also be able to use
	system calls such as exec or spawn (or execl, execv, spawnl,
	spawnv, etc.).

	See also question 19.30.

	References: K&R1 Sec. 7.9 p. 157; K&R2 Sec. 7.8.4 p. 167,
	Sec. B6 p. 253; ISO Sec. 7.10.4.5; H&S Sec. 19.2 p. 407; PCS
	Sec. 11 p. 179.

19.30:	How can I invoke another program or command and trap its output?

A:	Unix and some other systems provide a popen() function, which
	sets up a stdio stream on a pipe connected to the process
	running a command, so that the output can be read (or the input
	supplied).  (Also, remember to call pclose() when you're done.)

	If you can't use popen(), you may be able to use system(), with
	the output going to a file which you then open and read.

	If you're using Unix and popen() isn't sufficient, you can learn
	about pipe(), dup(), fork(), and exec().

	(One thing that probably would *not* work, by the way, would be
	to use freopen().)

	References: PCS Sec. 11 p. 169.

19.31:	How can my program discover the complete pathname to the
	executable from which it was invoked?

A:	argv[0] may contain all or part of the pathname, or it may
	contain nothing.  You may be able to duplicate the command
	language interpreter's search path logic to locate the
	executable if the name in argv[0] is present but incomplete.
	However, there is no guaranteed solution.

	References: K&R1 Sec. 5.11 p. 111; K&R2 Sec. 5.10 p. 115; ISO
	Sec. 5.1.2.2.1; H&S Sec. 20.1 p. 416.

19.32:	How can I automatically locate a program's configuration files
	in the same directory as the executable?

A:	It's hard; see also question 19.31 above.  Even if you can
	figure out a workable way to do it, you might want to consider
	making the program's auxiliary (library) directory configurable,
	perhaps with an environment variable.  (It's especially
	important to allow variable placement of a program's
	configuration files when the program will be used by several
	people, e.g. on a multiuser system.)

19.33:	How can a process change an environment variable in its caller?

A:	It may or may not be possible to do so at all.  Different
	operating systems implement global name/value functionality
	similar to the Unix environment in different ways.  Whether the
	"environment" can be usefully altered by a running program, and
	if so, how, is system-dependent.

	Under Unix, a process can modify its own environment (some
	systems provide setenv() or putenv() functions for the purpose),
	and the modified environment is generally passed on to child
	processes, but it is *not* propagated back to the parent
	process.  Under MS-DOS, it's possible to manipulate the master
	copy of the environment, but the required techniques are arcane.
	(See an MS-DOS FAQ list.)

19.36:	How can I read in an object file and jump to locations in it?

A:	You want a dynamic linker or loader.  It may be possible to
	malloc some space and read in object files, but you have to know
	an awful lot about object file formats, relocation, etc.  Under
	BSD Unix, you could use system() and ld -A to do the linking for
	you.  Many versions of SunOS and System V have the -ldl library
	which allows object files to be dynamically loaded.  Under VMS,
	use LIB$FIND_IMAGE_SYMBOL.  GNU has a package called "dld".  See
	also question 15.13.

19.37:	How can I implement a delay, or time a user's response,
	with sub-second resolution?

A:	Unfortunately, there is no portable way.  Routines you might
	look for on your system include clock(), delay(), ftime(),
	gettimeofday(), msleep(), nap(), napms(), nanosleep(),
	setitimer(), sleep(), Sleep(), times(), and usleep().
	(A function called wait(), however, is at least under Unix *not*
	what you want.)  The select() and poll() calls (if available)
	can be pressed into service to implement simple delays.
	On MS-DOS machines, it is possible to reprogram the system timer
	and timer interrupts.

	Of these, only clock() is part of the ANSI Standard.  The
	difference between two calls to clock() gives elapsed execution
	time, and may even have subsecond resolution, if CLOCKS_PER_SEC
	is greater than 1.  However, clock() gives elapsed processor
	time used by the current program, which on a multitasking system
	may differ considerably from real time.

	If you're trying to implement a delay and all you have available
	is a time-reporting function, you can implement a CPU-intensive
	busy-wait, but this is only an option on a single-user, single-
	tasking machine, as it is terribly antisocial to any other
	processes.  Under a multitasking operating system, be sure to
	use a call which puts your process to sleep for the duration,
	such as sleep() or select(), or pause() in conjunction with
	alarm() or setitimer().

	For really brief delays, it's tempting to use a do-nothing loop
	like

		long int i;
		for(i = 0; i < 1000000; i++)
			;

	but resist this temptation if at all possible!  For one thing,
	your carefully-calculated delay loops will stop working properly
	next month when a faster processor comes out.  Perhaps worse, a
	clever compiler may notice that the loop does nothing and
	optimize it away completely.

	References: H&S Sec. 18.1 pp. 398-9; PCS Sec. 12 pp.
	197-8,215-6; POSIX Sec. 4.5.2.

19.38:	How can I trap or ignore keyboard interrupts like control-C?

A:	The basic step is to call signal(), either as

		#include 
		signal(SIGINT, SIG_IGN);

	to ignore the interrupt signal, or as

		extern void func(int);
		signal(SIGINT, func);

	to cause control to transfer to function func() on receipt of an
	interrupt signal.

	On a multi-tasking system such as Unix, it's best to use a
	slightly more involved technique:

		extern void func(int);
		if(signal(SIGINT, SIG_IGN) != SIG_IGN)
			signal(SIGINT, func);

	The test and extra call ensure that a keyboard interrupt typed
	in the foreground won't inadvertently interrupt a program
	running in the background (and it doesn't hurt to code calls to
	signal() this way on any system).

	On some systems, keyboard interrupt handling is also a function
	of the mode of the terminal-input subsystem; see question 19.1.
	On some systems, checking for keyboard interrupts is only
	performed when the program is reading input, and keyboard
	interrupt handling may therefore depend on which input routines
	are being called (and *whether* any input routines are active at
	all).  On MS-DOS systems, setcbrk() or ctrlbrk() functions may
	also be involved.

	References: ISO Secs. 7.7,7.7.1; H&S Sec. 19.6 pp. 411-3; PCS
	Sec. 12 pp. 210-2; POSIX Secs. 3.3.1,3.3.4.

19.39:	How can I handle floating-point exceptions gracefully?

A:	On many systems, you can define a function matherr() which will
	be called when there are certain floating-point errors, such as
	errors in the math routines in .  You may also be able
	to use signal() (see question 19.38 above) to catch SIGFPE.  See
	also question 14.9.

	References: Rationale Sec. 4.5.1.

19.40:	How do I...  Use sockets?  Do networking?  Write client/server
	applications?

A:	All of these questions are outside of the scope of this list and
	have much more to do with the networking facilities which you
	have available than they do with C.  Good books on the subject
	are Douglas Comer's three-volume _Internetworking with TCP/IP_
	and W. R. Stevens's _UNIX Network Programming_.  There is also
	plenty of information out on the net itself, including the
	"Unix Socket FAQ" at http://www.developerweb.net/sock-faq/ ,
	and "Beej's Guide to Network Programming" at
	http://www.ecst.csuchico.edu/~beej/guide/net/.

	(One tip: depending on your OS, you may need to explicitly
	request the -lsocket and -lnsl libraries; see question 13.25.)

19.40b:	How do I...  Use BIOS calls?  Write ISR's?  Create TSR's?

A:	These are very particular to specific systems (PC compatibles
	running MS-DOS, most likely).  You'll get much better
	information in a specific newsgroup such as
	comp.os.msdos.programmer or its FAQ list; another excellent
	resource is Ralf Brown's interrupt list.

19.40c:	I'm trying to compile this program, but the compiler is
	complaining that "union REGS" is undefined, and the linker
	is complaining that int86() is undefined.

A:	Those have to do with MS-DOS interrupt programming.  They don't
	exist on other systems.

19.40d:	What are "near" and "far" pointers?

A:	These days, they're pretty much obsolete; they're definitely
	system-specific.  If you really need to know, see a DOS- or
	Windows-specific programming reference.

19.41:	But I can't use all these nonstandard, system-dependent
	functions, because my program has to be ANSI compatible!

A:	You're out of luck.  Either you misunderstood your requirement,
	or it's an impossible one to meet.  ANSI/ISO Standard C simply
	does not define ways of doing these things; it is a language
	standard, not an operating system standard.  An international
	standard which does address many of these issues is POSIX
	(IEEE 1003.1, ISO/IEC 9945-1), and many operating systems (not
	just Unix) now have POSIX-compatible programming interfaces.

	It is possible, and desirable, for *most* of a program to be
	ANSI-compatible, deferring the system-dependent functionality to
	a few routines in a few files which are either heavily #ifdeffed
	or rewritten entirely for each system ported to.


Section 20. Miscellaneous

20.1:	How can I return multiple values from a function?

A:	Either pass pointers to several locations which the function can
	fill in, or have the function return a structure containing the
	desired values, or (in a pinch) you could theoretically use
	global variables.  See also questions 4.8 and 7.5a.

20.3:	How do I access command-line arguments?

A:	They are pointed to by the argv array with which main() is
	called.  See also questions 8.2, 13.7, and 19.20.

	References: K&R1 Sec. 5.11 pp. 110-114; K&R2 Sec. 5.10 pp.
	114-118; ISO Sec. 5.1.2.2.1; H&S Sec. 20.1 p. 416; PCS Sec. 5.6
	pp. 81-2, Sec. 11 p. 159, pp. 339-40 Appendix F; Schumacher,
	ed., _Software Solutions in C_ Sec. 4 pp. 75-85.

20.5:	How can I write data files which can be read on other machines
	with different word size, byte order, or floating point formats?

A:	The most portable solution is to use text files (usually ASCII),
	written with fprintf() and read with fscanf() or the like.
	(Similar advice also applies to network protocols.)  Be
	skeptical of arguments which imply that text files are too big,
	or that reading and writing them is too slow.  Not only is their
	efficiency frequently acceptable in practice, but the advantages
	of being able to interchange them easily between machines, and
	manipulate them with standard tools, can be overwhelming.

	If you must use a binary format, you can improve portability,
	and perhaps take advantage of prewritten I/O libraries, by
	making use of standardized formats such as Sun's XDR (RFC 1014),
	OSI's ASN.1 (referenced in CCITT X.409 and ISO 8825 "Basic
	Encoding Rules"), CDF, netCDF, or HDF.  See also questions 2.12
	and 12.38.

	References: PCS Sec. 6 pp. 86, 88.

20.6:	If I have a char * variable pointing to the name of a function,
	how can I call that function?

A:	The most straightforward thing to do is to maintain a
	correspondence table of names and function pointers:

		int one_func(), two_func();
		int red_func(), blue_func();

		struct { char *name; int (*funcptr)(); } symtab[] = {
			"one_func",	one_func,
			"two_func",	two_func,
			"red_func",	red_func,
			"blue_func",	blue_func,
		};

	Then, search the table for the name, and call via the associated
	function pointer.  See also questions 2.15, 18.14, and 19.36.

	References: PCS Sec. 11 p. 168.

20.8:	How can I implement sets or arrays of bits?

A:	Use arrays of char or int, with a few macros to access the
	desired bit at the proper index.  Here are some simple macros to
	use with arrays of char:

		#include 		/* for CHAR_BIT */

		#define BITMASK(b) (1 << ((b) % CHAR_BIT))
		#define BITSLOT(b) ((b) / CHAR_BIT)
		#define BITSET(a, b) ((a)[BITSLOT(b)] |= BITMASK(b))
		#define BITTEST(a, b) ((a)[BITSLOT(b)] & BITMASK(b))

	(If you don't have , try using 8 for CHAR_BIT.)

	References: H&S Sec. 7.6.7 pp. 211-216.

20.9:	How can I determine whether a machine's byte order is big-endian
	or little-endian?

A:	One way is to use a pointer:

		int x = 1;
		if(*(char *)&x == 1)
			printf("little-endian/n");
		else	printf("big-endian/n");

	It's also possible to use a union.

	See also questions 10.16 and 20.9b.

	References: H&S Sec. 6.1.2 pp. 163-4.

20.9b:	How do I swap bytes?

A:	V7 Unix had a swab() function, but it seems to have been
	forgotten.

	A problem with explicit byte-swapping code is that you have
	to decide whether to call it or not; see question 20.9 above.
	A better solution is to use functions (such as the BSD
	networking ntohs() et al.) which convert between the known byte
	order of the data and the (unknown) byte order of the machine,
	and to arrange for these functions to be no-ops on those
	machines which already match the desired byte order.

	If you do have to write your own byte-swapping code, the two
	obvious approaches are again to use pointers or unions, as in
	question 20.9.

	References: PCS Sec. 11 p. 179.

20.10:	How can I convert integers to binary or hexadecimal?

A:	Make sure you really know what you're asking.  Integers are
	stored internally in binary, although for most purposes it is
	not incorrect to think of them as being in octal, decimal, or
	hexadecimal, whichever is convenient.  The base in which a
	number is expressed matters only when that number is read in
	from or written out to the outside world.

	In source code, a non-decimal base is indicated by a leading 0
	or 0x (for octal or hexadecimal, respectively).  During I/O, the
	base of a formatted number is controlled in the printf and scanf
	family of functions by the choice of format specifier (%d, %o,
	%x, etc.) and in the strtol() and strtoul() functions by the
	third argument.  If you need to output numeric strings in
	arbitrary bases, you'll have to supply your own function to do
	so (it will essentially be the inverse of strtol).  During
	*binary* I/O, however, the base again becomes immaterial.

	For more information about "binary" I/O, see question 2.11.
	See also questions 8.6 and 13.1.

	References: ISO Secs. 7.10.1.5,7.10.1.6.

20.11:	Can I use base-2 constants (something like 0b101010)?
	Is there a printf() format for binary?

A:	No, on both counts.  You can convert base-2 string
	representations to integers with strtol().  See also question
	20.10.

20.12:	What is the most efficient way to count the number of bits which
	are set in an integer?

A:	Many "bit-fiddling" problems like this one can be sped up and
	streamlined using lookup tables (but see question 20.13 below).

20.13:	What's the best way of making my program efficient?

A:	By picking good algorithms, implementing them carefully, and
	making sure that your program isn't doing any extra work.  For
	example, the most microoptimized character-copying loop in the
	world will be beat by code which avoids having to copy
	characters at all.

	When worrying about efficiency, it's important to keep several
	things in perspective.  First of all, although efficiency is an
	enormously popular topic, it is not always as important as
	people tend to think it is.  Most of the code in most programs
	is not time-critical.  When code is not time-critical, it is
	usually more important that it be written clearly and portably
	than that it be written maximally efficiently.  (Remember that
	computers are very, very fast, and that seemingly "inefficient"
	code may be quite efficiently compilable, and run without
	apparent delay.)

	It is notoriously difficult to predict what the "hot spots" in a
	program will be.  When efficiency is a concern, it is important
	to use profiling software to determine which parts of the
	program deserve attention.  Often, actual computation time is
	swamped by peripheral tasks such as I/O and memory allocation,
	which can be sped up by using buffering and caching techniques.

	Even for code that *is* time-critical, one of the least
	effective optimization techniques is to fuss with the coding
	details.  Many of the "efficient coding tricks" which are
	frequently suggested (e.g. substituting shift operators for
	multiplication by powers of two) are performed automatically by
	even simpleminded compilers.  Heavyhanded optimization attempts
	can make code so bulky that performance is actually degraded,
	and are rarely portable (i.e. they may speed things up on one
	machine but slow them down on another).  In any case, tweaking
	the coding usually results in at best linear performance
	improvements; the big payoffs are in better algorithms.

	For more discussion of efficiency tradeoffs, as well as good
	advice on how to improve efficiency when it is important, see
	chapter 7 of Kernighan and Plauger's _The Elements of
	Programming Style_, and Jon Bentley's _Writing Efficient
	Programs_.

20.14:	Are pointers really faster than arrays?  How much do function
	calls slow things down?  Is ++i faster than i = i + 1?

A:	Precise answers to these and many similar questions depend of
	course on the processor and compiler in use.  If you simply must
	know, you'll have to time test programs carefully.  (Often the
	differences are so slight that hundreds of thousands of
	iterations are required even to see them.  Check the compiler's
	assembly language output, if available, to see if two purported
	alternatives aren't compiled identically.)

	For conventional machines, it is usually faster to march through
	large arrays with pointers rather than array subscripts, but for
	some processors the reverse is true.

	Function calls, though obviously incrementally slower than in-
	line code, contribute so much to modularity and code clarity
	that there is rarely good reason to avoid them.

	Before rearranging expressions such as i = i + 1, remember that
	you are dealing with a compiler, not a keystroke-programmable
	calculator.  Any decent compiler will generate identical code
	for ++i, i += 1, and i = i + 1.  The reasons for using ++i or
	i += 1 over i = i + 1 have to do with style, not efficiency.
	(See also question 3.12b.)

20.15b:	People claim that optimizing compilers are good and that we no
	longer have to write things in assembler for speed, but my
	compiler can't even replace i/=2 with a shift.

A:	Was i signed or unsigned?  If it was signed, a shift is not
	equivalent (hint: think about the result if i is negative and
	odd), so the compiler was correct not to use it.

20.15c:	How can I swap two values without using a temporary?

A:	The standard hoary old assembly language programmer's trick is:

		a ^= b;
		b ^= a;
		a ^= b;

	But this sort of code has little place in modern, HLL
	programming.  Temporary variables are essentially free,
	and the idiomatic code using three assignments, namely

		int t = a;
		a = b;
		b = t;

	is not only clearer to the human reader, it is more likely to be
	recognized by the compiler and turned into the most-efficient
	code (e.g. perhaps even using an EXCH instruction).  The latter
	code is obviously also amenable to use with pointers and
	floating-point values, unlike the XOR trick.  See also questions
	3.3b and 10.3.

20.17:	Is there a way to switch on strings?

A:	Not directly.  Sometimes, it's appropriate to use a separate
	function to map strings to integer codes, and then switch on
	those.  Otherwise, of course, you can fall back on strcmp() and
	a conventional if/else chain.  See also questions 10.12, 20.18,
	and 20.29.

	References: K&R1 Sec. 3.4 p. 55; K&R2 Sec. 3.4 p. 58; ISO
	Sec. 6.6.4.2; H&S Sec. 8.7 p. 248.

20.18:	Is there a way to have non-constant case labels (i.e. ranges or
	arbitrary expressions)?

A:	No.  The switch statement was originally designed to be quite
	simple for the compiler to translate, therefore case labels are
	limited to single, constant, integral expressions.  You *can*
	attach several case labels to the same statement, which will let
	you cover a small range if you don't mind listing all cases
	explicitly.

	If you want to select on arbitrary ranges or non-constant
	expressions, you'll have to use an if/else chain.

	See also question 20.17.

	References: K&R1 Sec. 3.4 p. 55; K&R2 Sec. 3.4 p. 58; ISO
	Sec. 6.6.4.2; Rationale Sec. 3.6.4.2; H&S Sec. 8.7 p. 248.

20.19:	Are the outer parentheses in return statements really optional?

A:	Yes.

	Long ago, in the early days of C, they were required, and just
	enough people learned C then, and wrote code which is still in
	circulation, that the notion that they might still be required
	is widespread.

	(As it happens, parentheses are optional with the sizeof
	operator, too, under certain circumstances.)

	References: K&R1 Sec. A18.3 p. 218; ISO Sec. 6.3.3, Sec. 6.6.6;
	H&S Sec. 8.9 p. 254.

20.20:	Why don't C comments nest?  How am I supposed to comment out
	code containing comments?  Are comments legal inside quoted
	strings?

A:	C comments don't nest mostly because PL/I's comments, which C's
	are borrowed from, don't either.  Therefore, it is usually
	better to "comment out" large sections of code, which might
	contain comments, with #ifdef or #if 0 (but see question 11.19).

	The character sequences /* and */ are not special within double-
	quoted strings, and do not therefore introduce comments, because
	a program (particularly one which is generating C code as
	output) might want to print them.

	Note also that // comments have only become legal in C as of
	C99.

	References: K&R1 Sec. A2.1 p. 179; K&R2 Sec. A2.2 p. 192; ISO
	Sec. 6.1.9, Annex F; Rationale Sec. 3.1.9; H&S Sec. 2.2 pp.
	18-9; PCS Sec. 10 p. 130.

20.21b:	Is C a great language, or what?  Where else could you write
	something like a+++++b ?

A:	Well, you can't meaningfully write it in C, either.
	The rule for lexical analysis is that at each point during a
	straightforward left-to-right scan, the longest possible token
	is determined, without regard to whether the resulting sequence
	of tokens makes sense.  The fragment in the question is
	therefore interpreted as

		a ++ ++ + b

	and cannot be parsed as a valid expression.

	References: K&R1 Sec. A2 p. 179; K&R2 Sec. A2.1 p. 192; ISO
	Sec. 6.1; H&S Sec. 2.3 pp. 19-20.

20.24:	Why doesn't C have nested functions?

A:	It's not trivial to implement nested functions such that they
	have the proper access to local variables in the containing
	function(s), so they were deliberately left out of C as a
	simplification.  (gcc does allow them, as an extension.)  For
	many potential uses of nested functions (e.g. qsort comparison
	functions), an adequate if slightly cumbersome solution is to
	use an adjacent function with static declaration, communicating
	if necessary via a few static variables.  (A cleaner solution,
	though unsupported by qsort(), is to pass around a pointer to
	a structure containing the necessary context.)

20.24b:	What is assert() and when would I use it?

A:	It is a macro, defined in , for testing "assertions".
	An assertion essentially documents an assumption being made by
	the programmer, an assumption which, if violated, would indicate
	a serious programming error.  For example, a function which was
	supposed to be called with a non-null pointer could write

		assert(p != NULL);

	A failed assertion terminates the program.  Assertions should
	*not* be used to catch expected errors, such as malloc() or
	fopen() failures.

	References: K&R2 Sec. B6 pp. 253-4; ISO Sec. 7.2; H&S Sec. 19.1
	p. 406.

20.25:	How can I call FORTRAN (C++, BASIC, Pascal, Ada, LISP) functions
	from C?  (And vice versa?)

A:	The answer is entirely dependent on the machine and the specific
	calling sequences of the various compilers in use, and may not
	be possible at all.  Read your compiler documentation very
	carefully; sometimes there is a "mixed-language programming
	guide," although the techniques for passing arguments and
	ensuring correct run-time startup are often arcane.

	For FORTRAN, more information may be found in FORT.gz by Glenn
	Geers, available via anonymous ftp from suphys.physics.su.oz.au
	in the src directory.  Burkhard Burow's header file cfortran.h
	simplifies C/FORTRAN interfacing on many popular machines.
	It is available via anonymous ftp from zebra.desy.de or at
	http://www-zeus.desy.de/~burow .

	In C++, a "C" modifier in an external function declaration
	indicates that the function is to be called using C calling
	conventions.

	References: H&S Sec. 4.9.8 pp. 106-7.

20.26:	Does anyone know of a program for converting Pascal or FORTRAN
	(or LISP, Ada, awk, "Old" C, ...) to C?

A:	Several freely distributable programs are available:

	p2c	A Pascal to C converter written by Dave Gillespie,
		posted to comp.sources.unix in March, 1990 (Volume 21);
		also available by anonymous ftp from
		csvax.cs.caltech.edu, file pub/p2c-1.20.tar.Z .

	ptoc	Another Pascal to C converter, this one written in
		Pascal (comp.sources.unix, Volume 10, also patches in
		Volume 13?).

	f2c	A FORTRAN to C converter jointly developed by people
		from Bell Labs, Bellcore, and Carnegie Mellon.  To find
		out more about f2c, send the mail message "send index
		from f2c" to [email protected] or research!netlib.
		(It is also available via anonymous ftp on
		netlib.att.com, in directory netlib/f2c/.)

	This FAQ list's maintainer also has available a list of a few
	other translators.

	See also questions 11.31 and 18.16.

20.27:	Is C++ a superset of C?  Can I use a C++ compiler to compile C
	code?

A:	C++ was derived from C, and is largely based on it, but there
	are some legal C constructs which are not legal C++.
	Conversely, ANSI C inherited several features from C++,
	including prototypes and const, so neither language is really a
	subset or superset of the other; the two also define the meaning
	of some common constructs differently.  In spite of the
	differences, many C programs will compile correctly in a C++
	environment, and many recent compilers offer both C and C++
	compilation modes.  (But it's usually a bad idea to compile
	straight C code as if it were C++; the languages are different
	enough that you'll generally get poor results.)  See also
	questions 8.9 and 20.20.

	References: H&S p. xviii, Sec. 1.1.5 p. 6, Sec. 2.8 pp. 36-7,
	Sec. 4.9 pp. 104-107.

20.28:	I need a sort of an "approximate" strcmp routine, for comparing
	two strings for close, but not necessarily exact, equality.

A:	Some nice information and algorithms having to do with
	approximate string matching, as well as a useful bibliography,
	can be found in Sun Wu and Udi Manber's paper "AGREP -- A Fast
	Approximate Pattern-Matching Tool."

	Another approach involves the "soundex" algorithm, which maps
	similar-sounding words to the same codes.  Soundex was designed
	for discovering similar-sounding names (for telephone directory
	assistance, as it happens), but it can be pressed into service
	for processing arbitrary words.

	References: Knuth Sec. 6 pp. 391-2 Volume 3; Wu and Manber,
	"AGREP -- A Fast Approximate Pattern-Matching Tool" .

20.29:	What is hashing?

A:	Hashing is the process of mapping strings to integers, usually
	in a relatively small range.  A "hash function" maps a string
	(or some other data structure) to a bounded number (the "hash
	bucket") which can more easily be used as an index in an array,
	or for performing repeated comparisons.  (Obviously, a mapping
	from a potentially huge set of strings to a small set of
	integers will not be unique.  Any algorithm using hashing
	therefore has to deal with the possibility of "collisions.")
	Many hashing functions and related algorithms have been
	developed; a full treatment is beyond the scope of this list.

	References: K&R2 Sec. 6.6; Knuth Sec. 6.4 pp. 506-549 Volume 3;
	Sedgewick Sec. 16 pp. 231-244.

20.31:	How can I find the day of the week given the date?

A:	Use mktime() or localtime() (see questions 13.13 and 13.14, but
	beware of DST adjustments if tm_hour is 0), or Zeller's
	congruence (see the sci.math FAQ list), or this elegant code by
	Tomohiko Sakamoto:

		int dayofweek(int y, int m, int d)	/* 0 = Sunday */
		{
			static int t[] = {0, 3, 2, 5, 0, 3, 5, 1, 4, 6, 2, 4};
			y -= m < 3;
			return (y + y/4 - y/100 + y/400 + t[m-1] + d) % 7;
		}

	See also questions 13.14 and 20.32.

	References: ISO Sec. 7.12.2.3.

20.32:	Is (year % 4 == 0) an accurate test for leap years?  (Was 2000 a
	leap year?)

A:	No, it's not accurate (and yes, 2000 was a leap year).
	The full expression for the present Gregorian calendar is

		year % 4 == 0 && (year % 100 != 0 || year % 400 == 0)

	See a good astronomical almanac or other reference for details.
	(To forestall an eternal debate: references which claim the
	existence of a 4000-year rule are wrong.)  See also question
	13.14.

20.34:	Here's a good puzzle: how do you write a program which produces
	its own source code as output?

A:	It is actually quite difficult to write a self-reproducing
	program that is truly portable, due particularly to quoting and
	character set difficulties.

	Here is a classic example (which ought to be presented on one
	line, although it will fix itself the first time it's run):

		char*s="char*s=%c%s%c;main(){printf(s,34,s,34);}";
		main(){printf(s,34,s,34);}

	(This program has a few deficiencies, among other things
	neglecting to #include , and assuming that the double-
	quote character " has the value 34, as it does in ASCII.)

	Here is an improved version, posted by James Hu:

		#define q(k)main(){return!puts(#k"/nq("#k")");}
		q(#define q(k)main(){return!puts(#k"/nq("#k")");})

20.35:	What is "Duff's Device"?

A:	It's a devastatingly devious way of unrolling a loop, devised by
	Tom Duff while he was at Lucasfilm.  In its "classic" form, it
	was used to copy bytes, and looked like this:

		register n = (count + 7) / 8;	/* count > 0 assumed */
		switch (count % 8)
		{
		case 0:    do {	*to = *from++;
		case 7:		*to = *from++;
		case 6:		*to = *from++;
		case 5:		*to = *from++;
		case 4:		*to = *from++;
		case 3:		*to = *from++;
		case 2:		*to = *from++;
		case 1:		*to = *from++;
			      } while (--n > 0);
		}

	where count bytes are to be copied from the array pointed to by
	from to the memory location pointed to by to (which is a memory-
	mapped device output register, which is why to isn't
	incremented).  It solves the problem of handling the leftover
	bytes (when count isn't a multiple of 8) by interleaving a
	switch statement with the loop which copies bytes 8 at a time.
	(Believe it or not, it *is* legal to have case labels buried
	within blocks nested in a switch statement like this.  In his
	announcement of the technique to C's developers and the world,
	Duff noted that C's switch syntax, in particular its "fall
	through" behavior, had long been controversial, and that "This
	code forms some sort of argument in that debate, but I'm not
	sure whether it's for or against.")

20.36:	When will the next International Obfuscated C Code Contest
	(IOCCC) be held?  How can I get a copy of the current and
	previous winning entries?

A:	The contest schedule varies over time; see
	http://www.ioccc.org/index.html for current details.

	Contest winners are usually announced at a Usenix conference,
	and are posted to the net sometime thereafter.  Winning entries
	from previous years (back to 1984) are archived at ftp.uu.net
	(see question 18.16) under the directory pub/ioccc/; see also
	http://www.ioccc.org/index.html .

20.37:	What was the entry keyword mentioned in K&R1?

A:	It was reserved to allow the possibility of having functions
	with multiple, differently-named entry points, a la FORTRAN.  It
	was not, to anyone's knowledge, ever implemented (nor does
	anyone remember what sort of syntax might have been imagined for
	it).  It has been withdrawn, and is not a keyword in ANSI C.
	(See also question 1.12.)

	References: K&R2 p. 259 Appendix C.

20.38:	Where does the name "C" come from, anyway?

A:	C was derived from Ken Thompson's experimental language B, which
	was inspired by Martin Richards's BCPL (Basic Combined
	Programming Language), which was a simplification of CPL
	(Combined Programming Language, or perhaps Cambridge Programming
	Language).  For a while, there was speculation that C's
	successor might be named P (the third letter in BCPL) instead of
	D, but of course the most visible descendant language today is C++.

20.39:	How do you pronounce "char"?

A:	You can pronounce the C keyword "char" in at least three ways:
	like the English words "char," "care," or "car" (or maybe even
	"character"); the choice is arbitrary.

20.39b:	What do "lvalue" and "rvalue" mean?

A:	Simply speaking, an "lvalue" is an expression that could appear
	on the left-hand sign of an assignment; you can also think of it
	as denoting an object that has a location.  (But see question
	6.7 concerning arrays.)  An "rvalue" is any expression that has
	a value (and that can therefore appear on the right-hand sign of
	an assignment).

20.40:	Where can I get extra copies of this list?

A:	An up-to-date copy may be obtained from ftp.eskimo.com in
	directory u/s/scs/C-faq/.  You can also just pull it off the
	net; it is normally posted to comp.lang.c on the first of each
	month, with an Expires: line which should keep it around all
	month.  A parallel, abridged version is available (and posted),
	as is a list of changes accompanying each significantly updated
	version.

	The various versions of this list are also posted to the
	newsgroups comp.answers and news.answers.  Several sites
	archive news.answers postings and other FAQ lists, including
	this one; two sites are rtfm.mit.edu (directories
	pub/usenet/news.answers/C-faq/ and pub/usenet/comp.lang.c/) and
	ftp.uu.net (directory usenet/news.answers/C-faq/).  If you don't
	have ftp access, a mailserver at rtfm.mit.edu can mail you FAQ
	lists: send a message containing the single word "help" to
	[email protected] .  See the meta-FAQ list in
	news.answers for more information.

	A hypertext (HTML) version of this FAQ list is available on the
	World-Wide Web; the URL is http://www.eskimo.com/~scs/C-faq/top.html .
	A comprehensive site which references all Usenet FAQ lists is
	http://www.faqs.org/faqs/ .

	An extended version of this FAQ list has been published by
	Addison-Wesley as _C Programming FAQs: Frequently Asked
	Questions_ (ISBN 0-201-84519-9).  An errata list is at
	http://www.eskimo.com/~scs/C-faq/book/Errata.html and on
	ftp.eskimo.com in u/s/scs/ftp/C-faq/book/Errata .

Bibliography

American National Standards Institute, _American National Standard for
Information Systems -- Programming Language -- C_, ANSI X3.159-1989
(see question 11.2).  [ANSI]

American National Standards Institute, _Rationale for American National
Standard for Information Systems -- Programming Language -- C_
(see question 11.2).  [Rationale]

Jon Bentley, _Writing Efficient Programs_, Prentice-Hall, 1982,
ISBN 0-13-970244-X.

David Burki, "Date Conversions," _The C Users Journal_, February 1993,
pp. 29-34.

Ian F. Darwin, _Checking C Programs with lint_, O'Reilly, 1988,
ISBN 0-937175-30-7.

David Goldberg, "What Every Computer Scientist Should Know about
Floating-Point Arithmetic," _ACM Computing Surveys_, Vol. 23 #1,
March, 1991, pp. 5-48.

Samuel P. Harbison and Guy L. Steele, Jr., _C: A Reference Manual_,
Fourth Edition, Prentice-Hall, 1995, ISBN 0-13-326224-3.  [There is
also a fifth edition: 2002, ISBN 0-13-089592-X.] [H&S]

Mark R. Horton, _Portable C Software_, Prentice Hall, 1990,
ISBN 0-13-868050-7.  [PCS]

Institute of Electrical and Electronics Engineers, _Portable Operating
System Interface (POSIX) -- Part 1: System Application Program Interface
(API) [C Language]_, IEEE Std. 1003.1, ISO/IEC 9945-1.

International Organization for Standardization, ISO 9899:1990
(see question 11.2).  [ISO]

International Organization for Standardization, WG14/N794 Working Draft
(see questions 11.1 and 11.2b).  [C9X]

Brian W. Kernighan and P.J. Plauger, _The Elements of Programming
Style_, Second Edition, McGraw-Hill, 1978, ISBN 0-07-034207-5.

Brian W. Kernighan and Dennis M. Ritchie, _The C Programming Language_,
Prentice-Hall, 1978, ISBN 0-13-110163-3.  [K&R1]

Brian W. Kernighan and Dennis M. Ritchie, _The C Programming Language_,
Second Edition, Prentice Hall, 1988, ISBN 0-13-110362-8, 0-13-110370-9.
(See also question 18.10.) [K&R2]

Donald E. Knuth, _The Art of Computer Programming_.  Volume 1:
_Fundamental Algorithms_, Third Edition, Addison-Wesley, 1997, ISBN
0-201-89683-4.  Volume 2: _Seminumerical Algorithms_, Third Edition,
1997, ISBN 0-201-89684-2.  Volume 3: _Sorting and Searching_, Second
Edition, 1998, ISBN 0-201-89685-0.  [Knuth]

Andrew Koenig, _C Traps and Pitfalls_, Addison-Wesley, 1989,
ISBN 0-201-17928-8.  [CT&P]

G. Marsaglia and T.A. Bray, "A Convenient Method for Generating Normal
Variables," _SIAM Review_, Vol. 6 #3, July, 1964.

Stephen K. Park and Keith W. Miller, "Random Number Generators: Good
Ones are Hard to Find," _Communications of the ACM_, Vol. 31 #10,
October, 1988, pp. 1192-1201 (also technical correspondence August,
1989, pp. 1020-1024, and July, 1993, pp. 108-110).

P.J. Plauger, _The Standard C Library_, Prentice Hall, 1992,
ISBN 0-13-131509-9.

Thomas Plum, _C Programming Guidelines_, Second Edition, Plum Hall,
1989, ISBN 0-911537-07-4.

William H. Press, Saul A. Teukolsky, William T. Vetterling, and Brian P.
Flannery, _Numerical Recipes in C_, Second Edition, Cambridge University
Press, 1992, ISBN 0-521-43108-5.

Dale Schumacher, Ed., _Software Solutions in C_, AP Professional, 1994,
ISBN 0-12-632360-7.

Robert Sedgewick, _Algorithms in C_, Addison-Wesley, 1990,
ISBN 0-201-51425-7.  (A new edition is being prepared; the first two
volumes are ISBN 0-201-31452-5 and 0-201-31663-3.)

Charles Simonyi and Martin Heller, "The Hungarian Revolution," _Byte_,
August, 1991, pp. 131-138.

David Straker, _C Style: Standards and Guidelines_, Prentice Hall,
ISBN 0-13-116898-3.

Steve Summit, _C Programming FAQs: Frequently Asked Questions_, Addison-
Wesley, 1995, ISBN 0-201-84519-9.  [The book version of this FAQ list;
see also http://www.eskimo.com/~scs/C-faq/book/Errata.html .]

Peter van der Linden, _Expert C Programming: Deep C Secrets_, Prentice
Hall, 1994, ISBN 0-13-177429-8.

Sun Wu and Udi Manber, "AGREP -- A Fast Approximate Pattern-Matching
Tool," USENIX Conference Proceedings, Winter, 1992, pp. 153-162.

There is another bibliography in the revised Indian Hill style guide
(see question 17.9).  See also question 18.10.